A SOCKS-BASED TECHNOLOGY FOR GAIT STATE ESTIMATION AND ALERT GENERATION USING EMBEDDED INERTIAL SENSORS PLACED ON THE METATARSAL SIDE

Abstract

Present invention pertaining to the socks-based technology 100 for estimating gait (or gait state) using embedded inertial sensors and generating alarms using actuators are disclosed. The socks comprised of inertial sensor 118 placed on the upper metatarsal side comprising of accelerometer and gyroscope, and actuators comprising of a speaker, vibrating motor, and a light emanating device such as LED 109 in PCB – array on dorsum 120 and pressure and force sensor on plantar side 121 is utilized for gait assessment and alert generation. Wherein the gait state for both the right and left leg is assessed using pre-filtering techniques with compensation for noise, drift, and gravity for further feature extraction through mathematical analysis. Features involving spatial, temporal, frequency, dynamic, and synchronic aspects are assessed using the Gait State Estimator (GSE) and orientation using the Pose State Estimator (PSE) using onboard computer 119 placed on the dorsum side 122. Upon exceeding the threshold limits of gait features, an abnormality is detected using an onboard computer, and an alarm is triggered using audio-visual devices to alert the person and/or the observer. FIG related to the abstract is FIG. 1.

Specification

Description:
Field of the Invention
Embodiments of the present invention relate to a socks based technology for estimating gait parameters using inertial sensors such as accelerometers and gyroscope sensors placed on the upper metatarsal side of the foot and alerting using audio-visual devices. The present invention also relates to developing a method for accurately estimating gait during the transition from one event to another in the time-varying signal of the acceleration and gyroscope data by following the gait phase sequences.


Background of the Invention
Measuring human movement parameters such as during normal walk has been a subject of immense interest to assess the nature of gait during normal and abnormal movement as during disability. Various technologies have been put forward for measuring human movement; examples include the use of 3-D motion capture systems along with force plates, magnetic systems, floor-mounted systems, and optoelectronic systems (Akhtaruzzaman M. D., 2016). Detection of several points using wireless motion capture at a short range is a well-known methodology applied in the medical diagnosis of foot abnormalities. In those applications mainly for detecting the movements of the digits, the marker size must be devoid of electrical connections for normal motion (Hashi S., 2006). Each marker is tracked by a minimum of three cameras at a time, where each camera provides a 2D view. Thus, a 3D data is reconstructed. However, due to the requirement for a large number of cameras, the setup is suitable only for indoor purposes. Using force plates along with the 3D motion capture system enables the recording of the Ground Reaction Forces (GRF). However, the data was restricted to straight-level walking only and did not capture measurements during various postures, which restricts the application of non-inertial gai measurements. Further, such a setup is highly expensive and not affordable for routine assessment in many laboratories, clinical centers, and hospitals.
The advent of wearable sensors enabled experimentation in diverse environments (indoor, outdoor, rough terrain). Biomedical sensors made of semiconductors and modern-day VLSI technology present a thrilling opportunity for measuring human physiological parameters in either continuous or real-time and also in a nonintrusive manner (Saw AE, 2015). These sensors are popularly known as Inertial Measurement Units (IMUs) and are an alternative to the expensive Gait analysis system (Seshadri DR, 2019). IMUs mainly comprise a tri-axial accelerometer and a tri-axial gyroscope used for the measurement of linear acceleration and angular velocity, respectively. The accelerometer senses the orientation by measuring the gravitational acceleration along its axes (Ahmed H, 2017). The sensor output, however, also measures the linear accelerations along with it and is corrupted due to the presence of Johnson-Nyquist Noise and Mechanical Noise.
Typically, in order to overcome the shortcomings of the sensor, the accelerometer measurements are passed through a filter such as the commonly used Kalman Filter (KF) to filter out inaccuracies in the data by minimizing the covariance error. KF is widely used because it can be established among all other possible linear filters, and this is the only one that diminishes the covariance of the estimation error. In notable research work, the algorithm has been able to detect the linear accelerations in the axes, and the measurement noise covariances have been increased in those corresponding axes only. This ultimately lowered the Kalman gain in those axes and preserved the important information of other axes (Ahmed H, 2017). Extended Kalman Filtering (EKF) is also utilized for drift error correction of the gyroscope. In experimental research, EKF gave an accuracy of over 97% for linear displacement (Bennett T, 2013). The proposed EKF method also proved to be more robust to ground height disturbances (Thatte N, 2019).
Further algorithms have been utilized unanimously to facilitate gait assessment. At a certain point around the midstance period of the Gait cycle, the foot lies flat on the floor, and the velocity and acceleration are measured to be zero. These zero velocity updates, when incorporated in the Kalman filter tuning, obtain a precise result for the position. This new algorithm is known as Zero Velocity Update (ZUPT). The effectiveness of ZUPT depends on the detection of zero velocity at the stance period (Bebek Ö, 2010).
Characterizing gait phase during each cycle is a challenge. Where, during normal walk, one can identify various time-based events during one gait cycle - TO, ISW, MSW, TSW, HS, FF, MST, TST, and HO. While gait assessment has been widely used for the analysis, further assessment of the plot, nature of wave form is not investigated, which reduces the scope of further characterization of the gait cycle in an individual. Further this also opens up precise gait variability assessment while examining successive gait cycle over time.
Despite numerous algorithms reported in the literature, specific details on individual gait patterns lack specific attention. Since uncertainty exists in the sensor output, accurate detection of gait events is essential for robust measurement. Where the gait patterns of the right and left leg may lose synchronism, there is an immense opportunity to explore these further aspects for applications in physiotherapy, neurorehabilitation, sports medicine, prosthesis, orthosis, bionics, and related areas. Since data collection is time-consuming in laboratory or clinical settings where ambulatory measurements are to be performed, such portable devices can aid in diagnosis and specialized administration of physiotherapy sessions or rehabilitation in real environments.


Summary of the Invention
The following is a brief summary of the subject matter that is described in greater detail herein. This summary is not intended to be limiting as to the scope of the claims.
It is an object of the invention to provide socks-based technology to assist in gait estimation employing inertial sensors on the metatarsal side where such positioning of the sensor has an advantage of capturing an event that coincides well with the results obtained using benchmark platforms such as 3D motion capture system.
It is an object of the invention to provide socks-based technology to facilitate the detection of key events of gait through the use of audio-visual-tactile devices to trigger an alarm or assist in alerting the subject or the observer. The socks have sensors and actuators that are controlled by the onboard computer, which performs mathematical calculations of pose and gait estimation.
It is an object of the invention to minimize the limitations in the prior art and help in the accurate and robust measurement of the gait, including features such as orientation, spatial, temporal, dynamic, frequency, synchronic, and related features for prospective identification of gait.
Another objective of the present invention is to design a network of interconnected inertial sensors to minimize the noise and increase the signal-to-noise ratio (SNR) for precise measurement of gait and pose parameters.
According to an aspect of the present invention, pressure and force sensors on the planar side, supported by a cushion, facilitate measurement of the forces of impact in relation to gait and pose.
In accordance with the aspect of the present invention, the fabric employs wires or conductive fibers to assist in the interconnectivity of the sensors, actuators, and peripherals to the onboard computer. The printed circuit board may be employed in totality or woven into fabrics as per suitability.
Finally, it is yet another objective of the present invention to facilitate ambulatory measurement by allowing for wireless charging of the battery via a battery management system (BMS) using coils that employ the technology of wireless power transfer.
These and other advantages and features of the present invention are described herein with specificity so as to make the present invention understandable to one of ordinary skill in the art.


Brief Description of the Drawings
Elements in the figures have not necessarily been drawn to scale; however, it improves the clarity and understanding of various elements of the embodiments of the present invention. Furthermore, elements that are known to be common and well understood to those in the industry are not depicted in order to provide a clear view of the various embodiments of the invention.
FIG. 1 represents the whole unit of the socks with onboard sensors and actuators, showing the front view, top view, and network connection of the sensor(s);
FIG. 2 is the time-varying signal output of the accelerometer sensor (showing acceleration along the X, Y, and Z-axis) before (top panel) and after filtering (bottom panel), where noise estimates are used to choose the tuning parameter to smoothen the time-varying signal optimally, without compromising the magnitude of the spike at TSW;
FIG. 3 describes the gait phases during walking;
FIG. 4 illustrates the gait phases of both the lower limbs during walking;
FIG. 8 depicts the flowchart for Gait State Estimator and Pose State Estimator
The drawings described herein are for illustration purposes only and are not intended to limit the scope of the present invention in any way.


Detailed Description of the Invention
The following description is merely exemplary in nature and is not intended to limit the present invention, applications, or uses. In the following discussion that addresses a number of embodiments and applications of the present invention, reference is made to the accompanying drawings that form a part hereof, where depictions are made, by way of illustration, of specific embodiments in which the invention may be practiced. It is to be understood that other embodiments may be utilized, and changes may be made without departing from the scope of the present invention.
According to an embodiment of the present invention, the socks comprise a plurality of sensors with necessary instrumentation and are employed for the acquisition of the sensor data along the X, Y, and Z axes. For example, the accelerometer is connected to an onboard microcontroller using a Serial Peripheral Interface (SPI) for wired communication. The sampling frequency may be chosen as 100 Hz (preferably 200 Hz for nominal recording and higher than 200 Hz for precise estimation) to ensure capturing gait events. Higher frequency setting of 3200 Hz may also be configured to perform precision estimation of the gait for clinical studies and during high-speed running, sprint and related activities.
According to an embodiment of the present invention, for high-speed sports, onboard microcontroller was employed to store the data in SRAM or SD-cards.
According to an embodiment of the present invention, for high-speed sports, remote acquisition of the data may also be established with wireless connectivity such as Zigbee, Bluetooth and WiFi.
In accordance with the embodiment of the present invention, the sensor node may be containing a plurality of sensor arranged in various combinations to allow for noise and gravity elimination.





In accordance with the embodiment of the present invention, descriptors of the gait may be defined to identify gait and pose features as mentioned in Table 1.
Table 1: Parameters of pose and gait
Feature type Features Definition/ Equation
Temporal Stride Time (ST)
Cadence (CDN)
Stance Time (STT)
Swing Time (SWT)
Single Support Time (SST)
Double Support Time (DST)
Gait cycle (GS)
Step Time (ST) ST = t(HS¬n+1)−t(HSn)
CDN = no of cycles/ minute
STT = t(TO)−t(HS)
SWT = t(HSn+1)−t(TOn)
SST = t(HSn+1)−t(TOn)
DST = t(TOn+1) −t(HSn)
GS = t(HSn+1) −t(HS) //t(TOn+1) −t(TO)
ST = t(HSn+1)−t(HSn)

Spatial Stride Length (STL)
Step Length (SL)
Step Width (SW)
Foot Angle (FA)
Walking speed (WS)
Ground reaction force (GRF) STL = p(HSn+1)−p(HSn)
SL = p(HSright)−p(HSleft)
SW = p(MHright)−p(MHleft)
FA = Orientation(foot) − Orientation (walking)
ie: O = Orientation
p = Position
Frequency Dominant Frequency Peak (DFP)
Dominant Frequency Peak BandWidth (BW-DFP)
BW-DFP= frequencymax - frequencymin
Synchronic Single Support Time(SST)
Double Support Time (DST)
Gait Asymmetry (GA)
Interpersonal Phase Difference (IPD)
Synchronisation Index (SI)
Bilateral Coordination (BC)
Inter-leg Coordination(ILC)
Phase Coordination Index (PCI) SST = t(HSn+1)−t(TOn)
DST = t(TOn+1)−t(HSn)
GA= ln(Right mean value (XR) / Left mean value (XL))×100%


Dynamic Ground Reaction Forces (GRF)
Anterior-Posterior Force (APF)
Medial-Lateral Force (MLF) GRF = Mass × Gravity

In reference to the method involving feature extraction of relevance to time, defined as Temporal Features, time related events are identified (FIG. 3). A gait cycle comprises of the Stance Phase (occupies 60% of the total gait cycle) and Swing Phase, which is further divided into sub-phases. The parameters of the stance phase comprise of Heel Strike (HS), Foot Flat (FF), Mid Stance (MST), Heel Off (HO) and Toe Off (TO). The Swing phase comprises of Initial Swing (ISW), Mid Swing (MSW) and Terminal Swing (TSW). Three peaks were notably distinguished in HL, TO and MSW. The time interval between HS and TO, TO and SW, SW and next HS are calculated for each Gait cycle. After that, it is observed whether the time interval remains constant throughout the Gait cycle and if any inference can be drawn from it.
Graphical representation of the data employs time on the X-axis and estimated acceleration on the Y-axis, offers an effective way to visually interpret temporal patterns and dynamic changes in a gait. Specifically related to the human gait cycle, a sequence of movements involved in walking, the graph highlights two pivotal phases -stance and swing phase. Where 60% of the gait cycle is made up of the stance phase, during which the foot touches the ground and provides necessary stability while the body weight is maintained. This comprises the Heel Strike (HS), the instant the foot touches the ground, initiating the first stage of dual support. It has the responsibility of making contact with the ground and starting weight acceptance. Foot Flat (FF): Starts at the point of initial contact and lasts until the opposing foot lifts off the ground. The foot continues to bear weight and absorb injuries by sliding towards pronation. Midstance (MST) initiates when the opposing foot lifts off the ground and lasts until the homolateral heel lifts from the ground. Here the supported of the body is aided by a single leg and begins to advance from force absorption in order to contact to force propulsion ahead. Heel off (HO) lasts from the moment the contralateral foot touches the ground until the heel comes off the floor. Toe Off (TO) phase begins as the toes leave the ground and extends till the contralateral foot touches the ground. Conversely, the swing phase, comprising 40%, depicts the duration when the foot is airborne, not in contact with the ground, as the leg propels forward for the subsequent step which is further split into further subphases: Initial swing lasts from the moment the toes leave the ground until the swinging foot is parallel to the stance foot. Mid-swing (MSW): This stage lasts until the swinging foot is precisely in front of the body, beginning when it is level with the stance foot. Terminal Swing (TSW) phase begins when the swinging foot is directly in front of the body and extends until the foot reaches the ground. Further granularity within the broader stance and swing phases reveals two specific support phases: Double Support-I, comprising (10-15% )facilitating stability during the transition between feet where both foot bears the body weight, and Single Support, comprising (35-40%) representing instances where only one foot bears the body weight in both the stance and swing phases. The time interval between HS and TO, TO and SW, SW and next HS are calculated for each Gait cycle. After that, it is observed whether the time interval remains constant throughout the Gait cycle and if any inference can be drawn from it.
In another reference graphical representation as shown in FIG. 3, displays an accurate overview of both the graphs representing the left and right legs gait phases. The red line signifies the right leg, while the black line signifies the left leg. The graph employs a notation scheme that starts with the highest peak of toe-off (TO) and ends with the next toe-off of the same foot, thereby representing a single gait cycle. This cycle includes both the stance and swing stages. The graphical representation precisely depicts the time-based changes in the walking cycle, where the vertical axis presumably reflects certain factors associated with the movement or location of the legs, and the horizontal axis indicates time. The apex of toe-off (TO) acts as a crucial landmark in the gait cycle, indicating the start of a new phase. The graph tracks the sequential occurrence of left and right foot toe-off events, enabling the observation of the phase difference between the two legs. The formula to quantify the phase difference between two successive left and right foot graphs is given by (PDF) = (tTO)Right - (tTO)Left. In this context, (tTO)Right refers to the specific moment when the right foot lifts off the ground, whereas (tTO)Left indicates the matching moment when the left foot does the same.
In reference to the method involving feature extraction of relevance to time, defined as Temporal Features, further features as defined as follows. Stride Time (ST): The time required to complete a full gait cycle, measured from the initial contact of foot to the following contact of the same foot. It gives a full assessment of an individual's walking rhythm and timing. Cadence (CDN) provides number of steps done over unit time. Stance Time (STT): The period each foot stays in contact with the ground throughout the gait cycle, indicating the weight-bearing phase. Crucial for analysing weight distribution, stability, and propulsion when walking. Swing Time (SWT): The period each foot spends off the ground throughout the gait cycle, showing the non-weight-bearing phase. It gives vital insights into the mechanics of limb movement. Double Support Time (DST): The interval during the locomotion cycle when both feet are concurrently in contact with the ground. In a full two step cycle, both feet are in touch with the ground at the same time for 20% of the entire gait cycle, with 10% in the beginning and conclusion of the stance phase. Single Support Time (SST): The period when just one foot carries the body's weight during the gait cycle. This temporal metric is crucial for analysing gait symmetry and balance. Gait Cycle (GC) : The full series of events from the first touch of one foot to the next contact of the same foot. An overall measure giving a complete view of the full walking pattern, often grouped into separate stance and swing stages for thorough study. Step Time (ST): The time necessary to perform one full step, ranging from the initial contact of one foot to the initial touch of the consecutive foot. This statistic refines the evaluation by concentrating on the time of each individual step.
In reference to the method involving feature extraction of relevance to space, defined as Spatial Features, are as follows. Stride Length (STL): A measure of the linear distance between two successive foot placements that sheds light on the overall distance traversed in a single gait cycle.The linear distance between one foot's first contact to the another foot's initial touch is commonly known as the step length (SL). The distance that each walking step covers is represented by this parameter. Step Width (SW): An essential measurement for assessing the lateral stability and base of support width while walking, this is the lateral distance measured between the midpoints of the heels of two successive foot placements. Foot Angle (FA): The angle created between the foot's longitudinal axis and the direction in which one is walking; this angle is useful for determining how the feet are oriented and aligned at various stages of the gait cycle. Walking Speed (WS): The speed at which a person moves through space when they are walking; this speed is often expressed in m/s or kph. a universal measure of functional ability and mobility.Walking-related Ground reaction force (GRF) is the force that the ground applies to the body. Comprehending the force distribution facilitates comprehension of the biomechanical and impact elements of every stride.
In reference to the method involving feature extraction of relevance to frequency, are as follows. The gait pattern's power spectral density (PSD) is often used to calculate the frequency parameters used in gait analysis. Dominant Frequency Peak (DFP) in Hz is the frequency in a signal's power spectrum with the greatest power. When it comes to gait analysis, the stride frequency or cadence often matches the main frequency peak. The most important repeating movement patterns during walking happen at this frequency. Applying a Fourier transform determines the frequency with the maximum magnitude, one can determine the dominant frequency peak. Dominant Frequency Peak Bandwidth (BW-DFP) in Hz: The range of frequencies where the frequency content's power is more than half of its. Whereas a narrower bandwidth would point to a more regular or steady gait, a wider bandwidth might imply a more varied gait.
In reference to the method involving feature extraction of relevance to dynamics and help in understanding the intricate mechanics of human movement are as follows. One such parameter is Ground Reaction Forces (GRF), which encompasses three key components. Firstly, the Vertical Force component represents the force exerted perpendicular to the ground during walking, with notable events such as heel strike and toe-off manifested in the vertical force curve. The component of GRF includes Anterior-Posterior Force (APF) which operates along the forward or backward direction of walking, providing insights into propulsion and braking forces. This measurement is instrumental in assessing gait efficiency and balance by capturing the intricate interplay of forces during locomotion. The second component, the Medial-Lateral Force (MLF) component acts laterally, signifying side-to-side movements and weight shifting. Monitoring changes in medial-lateral forces becomes crucial for evaluating gait stability and the control of balance. Together, these dynamic parameters offer a comprehensive perspective on the forces at play during walking, contributing valuable insights into the mechanics of human gait.
In reference to the method involving feature extraction of relevance to trace or trajectory aspects, are as follows. A gait trajectory in biomechanics is the three-dimensional course that a particular foot point such as the toe or heel follows throughout the swing and stance stages of the gait cycle. Understanding the trajectory of a foot requires analysing its movement in three different planes :- Sagittal Plane Trajectory: During the gait cycle, the foot follows a three-dimensional path in the forward-backward plane. It shows far and in which direction the foot moves on the front-to-back axis, which is vital information about forward motion and propulsion while walking. Frontal Plane Trajectory: This is the path taken by the foot as it moves side to side. It is crucial for evaluating walking stability and alignment and provides insightful information about lateral movements and deviations that might affect the dynamics of the gait as a whole. Transverse Plane Trajectory: This diagram shows the foot's three-dimensional journey in the rotating or up-down plane throughout the gait cycle. Understanding rotational motions, assessing overall foot mechanics, and gaining insight into walking coordination all depend on this trajectory. Whereas the components of Trajectory includes Foot Trajectory: The three-dimensional route that a foot point traces, reflecting the foot's movement in space throughout the gait cycle. Knee Trajectory: The route the knee takes while walking, which sheds light on the lower limb's mechanics. Hip Trajectory: The trajectory that describes how the hip joint moves and plays a role in the lower limb's overall coordination.
In reference to the method involving feature extraction of relevance to synchronism or synchronic aspects, are as follows. Single Support Time (SST): The amount of time throughout the gait cycle when just one foot carries the weight of the body. This temporal measurement is important to assess balance and symmetry in gait. The period of time during the gait cycle when both the feet are in contact with the ground at the same time is known as the Double Support Time (DST). A full two-step cycle has both feet touching the ground simultaneously for 20% of the gait cycle, 10% at the start of the stance phase, and 10% at the finish. Interpersonal phase difference (IPD) is the phase difference between two people's gait cycles. Synchronisation Index (SI): measures how well a person's gait mimics a stimulus from outside the body. SI is an acronym for statistical indicator of the connection between an individual's gait cycle and external inputs. Gait Asymmetry (GA): Gait Asymmetry refers to the variations between the right and left portions of the body during a walking endeavour. It can be measured through parameters such as stride length, stroke time, and stance time. Walking asymmetry means you take a slightly distinct stride with each limb. This may happen because of unequal leg lengths, different foot placements on one side, adjustments to the pelvic or back alignment, or just discomfort that affects your gait. GA = 0 signifies symmetry and GA >= 100% denotes asymmetry. Bilateral coordination (BC): It is the capacity of both sides of the body to coordinate at the same time . It describes the coordinated contraction of many muscles to synchronise the motions of the upper and lower limbs in the context of gait. Stronger frequency and phase locking between limbs indicate a higher level of interlimb coordination. Phase Coordination Index (PCI) : It is a temporal gait metric that assesses about the consistency and precision of the anti phased left-right pattern of stepping generation. It represents variability and inaccuracy, correspondingly, in phase generation. Numerous factors, including step length, body mass index, cadence, and gait speed, might influence PCI results. Gait Cycle (GCP): It is the cyclic pattern of movement that begins when the heel of one foot impacts the ground and ends when that same heel meets the ground once again. Inter-leg Coordination (ILC): Inter-leg Coordination refers to the temporal and spatial relationships between limb movements. Bilateral anti-phasing, or the beginning of one leg's swing period synchronised with the other leg's stance phase, is the fundamental basis of inter-leg synchronisation.
In accordance with the embodiment of the present invention concerning the proof of the concept, the time based events or temporal features are recognized and validated against 3D motion capture system. Events were identified on data sets involving 25 healthy subjects (19 male and 6 Female) who do not have any previous medical history of any cardiovascular, arthritis or any disability. All the subjects were told to do three different walking modes - slow, normal and fast walking of at least 70 -80 steps in each type. Their height, weight and age were recorded and BMI. (Body Mass Index) was calculated later. According to their BMI all the subjects can be classified in to Normal weighted and Over weighted individuals. The time intervals (Δt) between the peaks of Gait cycle of every subject for each style of walking were recorded. Analysis was done on that table to draw any inference.

Example 1: Sample recordings of the time event during walk
A tabular data comprising of the time points of HL, TO and SW along with Δt between HS - TO, TO - SW and SW - HS is being displayed below [ Fig 21].

Table 2: Data showing time ponts of the gait events recognized for a subject


Example 2: Inter and Intra-subject variability study for distinguishing normal and overweight persons
From the experimental data table average data of time intervals is found out. That average data in Table 02 helps us in distinguishing the datasets of normal and overweight persons.
Table 3: Time intervals of the gait events for one subject
SUB_ID STATUS WALK_STYLE STANCE
(sec) SWING
( sec) Stride
( sec) SD_Stance
( sec ) SD_Swing
(sec)
01. OW (BMI 26.6) SLOW 0.66
(56.4%) 0.51
(43.6%) 1.17 0.087 0.102
NORMAL 0.56
(56.6%) 0.43
(43.4%) 0.99 0.028 0.047
FAST 0.51
(56.7%) 0.39
(43.3%) 0.9 0.178 0.049
02. OW (BMI 27.9) SLOW 1.02
(60.2%) 0.675
(39.8%) 1.695 0.089 0.110
NORMAL 0.65
(56.5%) 0.5
(43.5%) 1.15 0.029 0.032
FAST 0.52
(55.3%) 0.42
(44.7%) 0.94 0.044 0.058
03. NW (BMI 22.3) SLOW 1.01
(64%) 0.571
(46%) 1.585 0.062 0.116
NORMAL 0.69
(57.1%) 0.52
(42.9%) 1.21 0.07 0.086
FAST 0.52
(55.2%) 0.423
(44.8%) 0.945 0.035 0.049
04. OW (BMI 25.3) SLOW 0.827
(56.2%) 0.644
(43.8%) 1.471 0.163 0.493
NORMAL 0.619
(57.6%) 0.455
(42.4%) 1.074 0.029 0.049
FAST 0.518
(54.7%) 0.429
(45.3%) 0.947 0.039 0.058
05. NW (BMI 24.1) SLOW 0.614
(57.2%) 0.46
(42.8%) 1.07 0.041 0.051
NORMAL 0.763
(57.4%) 0.566
(42.6%) 1.329 0.078 0.059
FAST 0.644
(56.9%) 0.487
(43.1%) 1.131 0.78 0.049
06. NW (BMI 20.4) SLOW 1.056
(60.5%) 0.69
(39.5%) 1.746 0.129 0.212
NORMAL 0.739
(57.6%) 0.545
(42.4%) 1.283 0.041 0.1
FAST 0.531
(53.7%) 0.458
(46.3%) 0.989 0.037 0.057
07. NW (BMI 21) SLOW 0.711
(51.5%) 0.671
(42.5%) 1.138 0.066 0.095
NORMAL 0.695
(56%) 0.546
(44%) 1.241 0.039 0.0556
FAST 0.541
(51.4%) 0.512
(48.6%) 1.053 0.061 0.074
08. OW (BMI 28) SLOW 0.705
(57.7%) 0.517
(42.3%) 1.22 0.042 0.057
NORMAL 0.699
(58%) 0.506
(42%) 1.205 0.148 0.038
FAST 0.563
(53.4%) 0.492
(46.6%) 1.055 0.045 0.066
09. OW (BMI 27.8) SLOW 0.684
(58.3%) 0.489
(41.7%) 1.173 0.017 0.076
NORMAL 0.728
(57.8%) 0.532
(42.3%) 1.26 0.057 0.067
FAST 0.52
(53.3%) 0.455
(46.7%) 0.975 0.031 0.041
10. NW (BMI 19.7) SLOW 1.036
(65%) 0.56
(35.1%) 1.59 0.098 0.146
NORMAL 0.697
(56.2%) 0.542
(43.8%) 1.239 0.07 0.127
FAST 0.498
(51.9%) 0.46
(48.1%) 0.958 0.036 0.051

Example 3: Comparison of time events for normal and overweight persons
The range of average stance time, swing time and stride time are studied for subjects of different BMI ranges for different walking patterns. The findings are listed below in Table 03.
Table 4: Temporal feature for walking styles for different BMI
Sl. No. Feature Walking Style Healthy weight
Feature in (sec) Over-weight
Feature in (sec)
01. STANCE TIME SLOW 0.614- 1.056
(51.4% - 64.9%) 0.66-1.02
(56.2% - 60.2%)
NORMAL 0.691 - 0.763
(56%- 57.6%) 0.56 - 0.73
(56.5%- 58.2%)
FAST 0.498 - 0.644
(51.4%- 56.9%) 0.51 - 0.56
(53.3% - 56.7%)
02. SWING TIME SLOW 0.46 - 0.69
(35.1% - 48.5%) 0.489 - 0.675
(39.8%-43.8%)
NORMAL 0.519 - 0.546
(42.5% - 44%) 0.43-0.53
(42%-43.5%)
FAST 0.423 - 0.512
(43.1% - 48.6%) 0.39-0.49
(43.3% - 46.6%)
03 STRIDE TIME SLOW 1.074 - 1.076 1.17-1.695
NORMAL 1.21 - 1.33 0.99 - 1.26
FAST 0.945 - 1.13 0.9-1.055

In accordance with the embodiment of the present invention, the gait parameters recognized can be used for diagnostic features.


In accordance with the embodiment of the present invention involving the algorithm, the flowchart involving the estimation of gait and pose features and the generation of alert signal as depitected in FIG. 5. Where, algorithm runs by following sequential processing of the discrete time varying signal during the run. The sensors showing accelerometer provides information on the acceleration over three mutually perpendicular axis and gyroscope provides information on the angular velocity about three mutually perpendicular axis. This is a raw data that bears noise and artifact errors arising from various reasons. To ensure artifacts are minimized, the sensor array 120 on the dorsum of the foot is firmly stick to the fabric by non-adhesive means, where the fabric underlying the sensor itself is held on to the skin underlying it by means of increased frictional resistance between the skin and the fabric. Alternatively the socks fabric may be loosely held at heel flap, turn, gusset, foot and toe with sticky like material covering the instep part of the socks. Such stickiness may hold the sensor array 120 to stay nearly motionless during normal ambulation.
In accordance to the present invention involving the algorithm, the sensor data is relayed to the onboard electronics 122 to perform further mathematical analysis. The onboard controller 119 acquires the data by wired communication (I2C, SPI, or other means) and stores the data in the memory which can be local or expanded to perform ambulatory recordings for few days. The raw data passes through a pre-filter which filters the noise by statistical means using KF/ EKF to smooth the plots, while ensuring the smoothening is sufficient enough to remove noise but does not affect the ampitude of the spike generated during TSW (FIG. 3).
With reference to pre-filtering, Kalman Filter employed on raw acceleration data show smoothening of the data with effective control over the extent of smoothness (FIG. 2). For this purpose, Kalman Filter equations were solved and testing on MATLAB platform. A for loop is run which accepts each measured value and provides the estimated correction to it according to the Kalman Gain. The Kalman Gain value depends on the covariance error. Greater the covariance error, lesser will be the Gain and the final estimate or correction will be nearer to the previous estimate or prediction. Similarly, lesser the covariance error, the corrected plot will follow the measured value. The obtained plot (FIG. 2 - lower panel) is more noise less than measured plot (FIG. 2 - upper panel). This obtained plot can hence be used in finding the essential time points of gait parameters and in peak detection. From this information we can study datasets of healthy subjects, overweight individuals and patients suffering from Parkinson diseases etc.
The current state update is evaluated by Kalman filtering using predicted state and sensor measurement values using the Kalman Gain factor. The Kalman Gain is calculated from covariance, which accounts for the uncertainty in predicting the system's state. Eventually, the new state formed has better estimated uncertainty than the previous one. The process runs in a loop, adjusting the Kalman Gain according to the covariance and moving towards a better state estimate
Kalman filtering algorithm is applicable for a linear system. The process and measurement model equations are,
a_t^- = F.a_(t-1)^+ + G.u_(t-1) + w_(t-1)
y_t = H.a_t + v_t
Where, a represents the state vector, y the measurement vector, F represents the state transition matrix, H representing observation matrix, v and w as white measurement and Gaussian process noises respectively.
The covariance matrix for process noise Q_(t-1) is depicted by:
Q_(t-1) = E(w_(t-1) w_(t-1)^T) , where E depicts the expectation operator.
Similarly, the covariance matrix describing the measurement noise R_t is defined as
R_t = E(v_t v_t^T)
Whereas KF perform linear approximation, to improve the accuracy Extend KF (EKF) is employed. The non-linear function at mean is evaluated by EKF as the best approximate of the distribution. Where, a slope is estimated around that mean. First order derivative of Taylor series expansion gives us the slope as a linear value is obtained from first order derivative only.
Let's say we have the following models of state transition and measurement:
a_t = f(a_(t-1),u_(t-1)) + w_(t-1)
y_t = h(a_t) + v_t
Where, a_t is the current state, f represents function of previous state a_(t-1) and control input u_(t-1). h is measurement function relating the current state a_t with the measurement y_t. w_(t-1) and v_t are process and measurement noises respectively, of covariances or uncertainty Q, and R respectively.
State Transition and Measurement Matrices obtained by performing Jacobian are represented as follows -
F_(t-1) = ∂f/∂a|(〖a^〗_(t-1)^+,u_(t-1))
H_t = ∂h/∂a| 〖a^〗_t^-
The hat "^" operator, means estimate of a variable. The superscripts "+" and "-" represent "a posteriori" and "a priori" respectively.
In accordance to the present invention involving the algorithm, further techniques are employed such as Zero-velocity Update (ZUPT) to facilite detection of key gait events which partly assist in drift correction and trajectory analysis. Hence upon successful elimination of drift error, gait phases of stance and swing can be accurately predicted.
In accordance to the present invention involving the algorithm, wherby, the process involving online calibration is performed using matching process involving sequence or pattern or correlation or statistical matching or those using AI/ML/DL approach.
Whereby the sequence based matching is described as follows. The stance phase detection is almost a sequence matching process, where the stance phase is observed under two circumstances -
Acceleration should be close to g, since the x- axis and y- axis of a calibrated accelerometer are 0 and that of z- axis is nearly 9.81.
Angular velocity will be 0, since for a calibrated gyroscope all the axis are 0.
The single detection threshold method is the conventional stance detection method of ZUPT algorithm. In this method, the IMU acceleration (a_t) and angular velocities (w_t) are compared with the ZUPT acceleration threshold (σa_t) and ZUPT angular velocity threshold (σw_t) respectively.
A stance phase is detected when,
a_t≤ σa_t and w_t≤ σw_t
Threshold gives the primary information of the entire ZUPT method. The threshold value of each activity like slow walking, fast walking, running are derived by summarizing the gait data changes from collected walking data.
Different experimental threshold values obtained from IMU sensors are as follows -
Normal walking range 〖 a〗_max - a_min < 3g
Fast walking range 3g < 〖 a〗_max - a_min < 4g
Range of climbing stairs 4g < 〖 a〗_max - a_min < 7g
Range of striding/jumping 〖 a〗_max - a_min > 7g
Where, 〖 a〗_max and a_min represent the magnitude of maximum acceleration and minimum acceleration respectively. g represents the gravitational acceleration. The angular threshold value (σw_t) is 0.6 rad/s. If the angular velocity (w_t) is less than 0.6 rad/s, stance phase will be represented. The disadvantage of the ZUPT algorithm is that it only utilises single threshold method for stance detection which is not an efficient way. The zero-velocity is sometimes detected prior to its occurrence; sometimes it misses to detect the phase. Due to these reasons advanced ZUPT stance detectors like double threshold method for stance detection are used nowadays.
In accordance to the present invention involving the algorithm, further techniques may be employed to derive best statistical estimates from KF, EKF combined with other statistical approaches for noise suppression/ elimination, drift correction and gravity compensation.
In accordance to the present invention involving the algorithm employing KF/EKF, state space of the gait and post gait and pose may be determined. Where, the estimates are run and updated over each gait cycle so the best estimate may be made.
In accordance to the present invention involving the algorithm employing noise suppression, drft correction and gravity compensation, the inertial sensor may be arranged in a network of 2, 3, and 4 to allow for improved SNR. The module employs sensors fixed at a distance of 'l' on the rod that is free to move relative to a reference coordinate of interest. For example, when using the accelerometer, it measures the components of the acceleration along the three perpendicular directions with respect to the local coordinate system. According to Newton's law of motion, accelerations can be represented mathematically for the system with two sensors as follows:

where, r0 is the position vector of the centre of the rod, and g is the acceleration due to gravity. Taking the difference of the acceleration from these two sensors cancels out the influence of the movement of the rod centroid and of the gravity, thus retaining information only about the changes of the vector l. The second derivative of the vector l is given by,

where, φ is the angle between the axes x (perpendicular to the vector l) and x', ω and α are the absolute angular velocity and angular acceleration of the rod, respectively. The difference in the acceleration in the direction along the rod axis depends on the square of the angular velocity and the angular acceleration. The proportionality coefficient is equal to the distance between the centers of the accelerometers. The two sensors eliminated the problem of gravity component and noise resulting from the signal. Now that the issues of noise were minimized using the two-sensor system, we can focus on the estimation of the joint angles and reduction of the drift with further analysis such as digital filtering, biased and statistical means for noise elimination.
In accordance with the embodiment of the present invention, the further features may be extracted using AI/ML/DL techniques as per suitability. Where, the features are futher assessed as those arising from right or left foot. The RL-LL estimates of gait state are derived and defined collectively as features of GSE, and those assessed through mathematical analysis of filtered inertial sensor data are defined collectively as features of PSE
In accordance with the embodiment of the present invention, the alert signals may be generated based on abnormal detection in gait and used as input to another system for possible hardware emulation purposes or general notification. For example in case of ambulatory usage by a subject, the subject may be informed of the possible abnormal walking pattern during real time by producing a sound such as a suitable ring tone using the speaker.
In accordance with the embodiment of the present invention, the alert signals may also utilize the sensory information from orther sensors such as pressure or force sensor, electromyography of foot associated muscles, and neural signal from related tissues of the foot.

REFERENCES - PATENT SOURCES
Mestrovic, Michael Anthony, et al. "System, garment and method." US Patent No. 9,186,092. 17 Nov. 2015.
Esposito, Mario, et al. "Sensor systems for user-specific evaluation of gait, footwear and garment fitting; monitoring of contact, force, pressure and/or shear at or near body surfaces." US Patent No. 11,154,243. 26 Oct. 2021.
Macagno, Maurizio, et al. "Sensor-enabled footwear; sensors, interfaces and sensor systems for data collection." US Patent Application No. 16/067,999.
Reif, Roberto, et al. "Sensor assemblies; sensor-enabled garments and objects; devices and systems for data collection." US Patent No. 11,060,926. 13 Jul. 2021.
Mariani, Benoît, and Kamiar Aminian. "System and method for 3D gait assessment." US Patent No. 9,307,932. 12 Apr. 2016.
Yang, Chang-Ming, et al. "System and method for analyzing gait using fabric sensors." US Patent No. 8,961,439. 24 Feb. 2015.

REFERENCES - NON-PATENT SOURCES
Akhtaruzzaman, M. D., Shafie, A. A., & Khan, M. R. (2016). Gait analysis: Systems, technologies, and importance. Journal of Mechanics in Medicine and Biology, 16(07), 1630003.
Hashi, S., Toyoda, M., Yabukami, S., Ishiyama, K., Okazaki, Y., & Arai, K. I. (2006). Wireless magnetic motion capture system for multi-marker detection. IEEE transactions on magnetics, 42(10), 3279-3281.
Saw, A. E., Main, L. C., & Gastin, P. B. (2015). Monitoring athletes through self-report: factors influencing implementation. Journal of sports science & medicine, 14(1), 137.
Seshadri, D. R., Li, R. T., Voos, J. E., Rowbottom, J. R., Alfes, C. M., Zorman, C. A., & Drummond, C. K. (2019). Wearable sensors for monitoring the physiological and biochemical profile of the athlete. NPJ digital medicine, 2(1), 72.
Ahmed, H., & Tahir, M. (2017). Improving the accuracy of human body orientation estimation with wearable IMU sensors. IEEE Transactions on instrumentation and measurement, 66(3), 535-542.
Bebek, Ö., Suster, M. A., Rajgopal, S., Fu, M. J., Huang, X., Çavuşoǧlu, M. C., ... & Mastrangelo, C. H. (2010). Personal navigation via high-resolution gait-corrected inertial measurement units. IEEE transactions on instrumentation and measurement, 59(11), 3018-3027.
Bennett, T., Jafari, R., & Gans, N. (2013, June). An extended kalman filter to estimate human gait parameters and walking distance. In 2013 American Control Conference (pp. 752-757). IEEE.
Thatte, N., Shah, T., & Geyer, H. (2019). Robust and adaptive lower limb prosthesis stance control via extended kalman filter-based gait phase estimation. IEEE Robotics and Automation Letters, 4(4), 3129-3136.

, C , Claims:I/WE claim
1. A system for estimating gait state and generating alert signals using fabric woven socks (100), comprising of -
inertial sensors at the Printed Circuit Board (PCB) - dorsum array (120) on the dorsum of the foot and comprising of accelerometer, gyroscope, and related inertial sensors;
actuators at the PCB - dorsum array (120) such as speaker (108), vibration device, light emitting diode (109), and related audio-visual-tactile devices that suit in accordance with the invention;
pressure or force sensor (112, 114, 116) on the dorsum of the foot;
battery with battery management system or BMS (110) to ensure power supply;
copper coil (111) for wireless charging of the battery via BMS;
the onboard computer (119), which performs data collection from the sensor, performs mathematical calculations in accordance with the underlying flowchart, and relays the essential information to the actuator for alerting through the use of peripherals in accordance with the invention;
wherein the memory module (local memory module/ expandable/ detachable type) serves to store the raw data or gait features;
the wireless device connects to the local device or server for remote communication, control, and data transfer operations and
features involving spatial, temporal, frequency, dynamic, and synchronic aspects are assessed using Gait State Estimator (GSE), and orientation and trace are assessed via Post State Estimator (PSE);
2. The socks (100) according to claim 1, wherein a combination of inertial sensors' (118) in one (101), two (102), and three (103, 104) may be arranged in a network to allow for compensation of noise, drift, and gravity.
3. The socks (100) according to claim 1, wherein the printed circuit (122) board housing the sensor/ actuator array, the computer, and inertial sensors on the dorsum are woven to the fabric and embedded in the fabric as a substrate;
4. The socks (100) according to claim 1, wherein onboard sensors and actuators and related peripheral devices may be arranged on the shank area (lateral or medial) (110), anterior, posterior (111), and dorsum (122) of the foot.
5. The socks (100) according to claim 1, wherein the onboard controller or computer is electrically connected to the PCB - shank (110), coil (111), pressure or force sensor (112, 114, 116), inertial sensors' (118), and PCB - dorsum array (120) on dorsum of the foot via copper wire or conductive fabric.
6. The socks (100) according to claim 1, wherein a coil (111) is placed at the posterior side of the foot for wireless charging.
7. The socks (100) according to claim 1, wherein a wireless module may be placed at PCB - shank (110) or one of an array of PCB - dorsum (111).
8. The socks (100) according to claim 1, wherein a detachable or expandable memory module may be placed at PCB - shank (110) or one of an array of PCB - dorsum (111), wherein the module connects to the said array by Velcro straps or non-adhesive straps.
9. The socks (100) according to claim 1, wherein the cuff (105) bears a reflective marker or coatings for easy identification by the camera for motion capture.
10. The socks (100) according to claim 1, wherein the sole bears an elastic or viscoelastic material to cushion the heel, mid-plantar, and toes'; where cushion support (113, 115, 117) lay next to the pressure or force sensor (112, 114, 116).
11. The socks (100) according to claim 1, wherein pre-filtering using Kalman Filter considers the choice of optimal values of tuning parameter estimated based on measured and estimated noise values where the degree of smoothening is sufficient enough for detection of peak at TSW.
12. The socks (100) according to claim 1, wherein the onboard or off-board mathematical calculation of the gait and post features is performed, wherein,
the underlying process involves suppression of noise, drift correction, and gravity compensation;
feature estimation for gait (using GSE) is assisted through the use of AI/ML/DL; and,
feature estimation for the pose is assisted through rigorous mathematical treatment where the pose estimates (using PSE) are derived through the use of data that has undergone pre-filtering, noise/ drift, and gravity compensation in addition to raw data from the inertial sensors.
13. The method, according to claim 12, wherein the temporal features of gait during one cycle are determined by capturing time points of the occurrence of events at TO, ISW, MSW, TSW, HS, FF, MST, TST, and HO.
14. The method according to claim 12, wherein the synchronic features of gait of right and left foot movement during one cycle are determined by correlating time points of the occurrence of events at TO, ISW, MSW, TSW, HS, FF, MST, TST, and HO.
15. The method according to claim 12, wherein the slope of the terrain and its roughness can be estimated through a change in the orientation of the curve during the stance phase.
16. The method according to claim 12, wherein the slope of the terrain and its roughness can be estimated through a change in the orientation of the curve leading to crest and trough at events - TO, ISW, MSW, TSW, HS, FF, MST, TST, and HO by following successive excursion of the gait cycle.
17. The method according to claim 12, wherein the nature of the terrain may be deduced as - flat, mesa, rough, rocky, slippery, loose and complex.
18. The method according to claim 12, wherein the occurrence of slip may be identified by measuring the time difference of TO of one foot and HS of another foot over successive excursion of the gait cycle.
19. The method according to claim 12, wherein peak acceleration and its change during foot contact to ground is measured for estimation of peak forces experienced during movement.
20. The method according to claim 12, wherein the pressure or forces sensors are placed beneath the foot and above the cushion support for assessment of the real forces of impact to the plantar.
21. The method according to claim 12, wherein variability in gait is estimated using subsequent time differences of the key gait events such as tTO, tISW, tMSW, tTSW, tHS, tFF, tMST, tTST, and tHO over successive excursion of the gait cycle.
22. The method according to claim 12, wherein variability is estimated by following drift in gait at events - TO, ISW, MSW, TSW, HS, FF, MST, TST, and HO over successive excursion of the gait cycle.
23. The method according to claim 12, wherein slope estimation of the acceleration during event transition over one cycle is made using a polynomial function, and further to this, its coefficients are employed for measurement of gait variability over successive excursion of the gait cycle.
24. The method according to claim 12, wherein calibration is performed in real-time and completes within a few gait cycles with a convergence of the error covariance matrix to within the tolerance.
25. The method according to claim 12, wherein flat foot estimation is determined during the event of MST of one foot while another foot is advancing from MSW to TSW.
26. The socks according to claim 12, wherein, for the measurement of time of impact and recoil during the course of TSW-HS transition, more samples are collected at a frequency greater than 100Hz.
27. The method according to claim 12, wherein orientation of the foot is determined by Pose State Estimator to trace the foot movement in three dimensions.

Documents