A SMART FOOTWEAR FOR DETECTING FALL OF A WEARER USING IOT AND MACHINE LEARNING TECHNIQUES

Abstract

ABSTRACT Disclosed herein is a smart footwear for detecting fall of a wearer. The smart footwear shown in FIG.1 comprises of a microcontroller (3) for processing data, communicating with IoT platform, and triggering alerts, a motion sensor module (2) for detecting motion, orientation, and changes in velocity of wearer, a sound sensor module (5) for detecting sound levels in vicinity of wearer, a GPS module (4) for enabling geo-fencing capabilities and allowing caregivers to define safe zones for wearer and a Li-ion battery (1) for powering smart footwear to ensure continuous operation throughout the day. The method for detecting fall comprises the steps of collecting data from various sensor modules, preparing collected data by standardizing sensor readings, selecting a model to analyze sensor data and training the model, using trained model to analyze incoming sensor data and triggering alerts for notifying family members of wearer and continuously improving accuracy and adaptation.

Specification

Description:FIELD OF THE INVENTION:

[0001] The present invention generally relates to fall detection systems. More particularly, the present invention relates to a smart footwear for senior citizens which detect their fall using IoT and machine learning algorithms.

BACKGROUND:

[0002] The rapid aging of global population has resulted in a growing need for affordable solutions to ensure the safety and well-being of people, especially senior citizens in indoor settings where falls can be a significant hazard. The advancement in technology has given rise to various devices for detecting falls of a person. Specifically, wearable devices such as smart watches and fitness trackers often include basic fall-detection capabilities. These devices use built-in accelerometers to detect sudden changes in motion that may indicate a fall. However, they are not specifically optimized for fall detection, leading to false alarms or missed detections. Also, they rely on the user to wear the device consistently, which is not always the case.

[0003] Some home monitoring systems utilize motion sensors and cameras to detect falls and monitor activity levels within the home. These monitoring systems provide a comprehensive view of the movements and behaviors of the individual. However, they are intrusive affecting the privacy concerns, as they involve constant video surveillance. Additionally, they require installation and maintenance, which is cumbersome and costly.

[0004] There are also mobile applications available that claim to offer fall detection functionality using the sensors built into smartphones. While these mobile applications are convenient and accessible, they lack accuracy and reliability, leading to false alarms or missed detections. Moreover, they rely on the individual to carry their smartphone at all times, which is not feasible, especially for those prone to falls. IoT-based systems integrate sensors placed throughout the home or wearable devices to monitor activity levels and detect falls. These systems provide real-time alerts to caregivers or emergency services when a fall is detected. However, they require complex installation and setup, as well as ongoing maintenance. Also, they are costly to implement and not accessible to all individuals, especially those from low-income backgrounds.

[0005] To overcome the aforementioned problems, there are many existing prior art which relates to detecting fall of a wearer. For example,

[0006] Indian Patent Application Number 202137060997 entitled "Fall Detection-Based Call-For-Help Method And Electronic Device" to Jiang, Yonghang and Chen, Xiaohan discloses a fall detection-based call-for-help method and an electronic device. The method is capable of improving the accuracy of electronic devices in fall detection and reduce the possibility of unintended trigger by the electronic devices for automatic call for help. The electronic device comprises of a motion sensor, and the motion sensor comprises an acceleration sensor and a gyroscope sensor. The electronic device acquires a first motion parameter of a user using the motion sensor. If the first motion parameter matches a first preset fall parameter, the electronic device obtains the fall confidence of the first motion parameter. The fall confidence of the first motion parameter is used for representing the degree of possibility that the first motion parameter is a motion parameter when the user falls. If the fall confidence of the first motion parameter is greater than a preset confidence threshold, the electronic device sends out call-for-help information.

[0007] A non-patent literature titled "A Real-time intelligent shoe system for fall detection" to Yanbo Tao et. al., relates to an intelligent shoe system which detect the fall and also classify the fall direction. The intelligent shoe comprises of eight pairs of force sensing resistors (FSRs) which are used to acquire forces in different location of the insole. To reduce the computational cost, power consumption, and enhance the real-time performance, the number of sensors is reduced based on principle component analysis (PCA) resulting in a four-pair version. An artificial neural network (ANN) classifies the system input into three observations, and develop a finite state machine to trigger correct alarm and prevent false alarm from other complex human actions.

[0008] The existing technologies relating to fall detection suffer from accuracy issues, leading to false alarms or missed detections. This can undermine trust in the system and lead to unnecessary stress for both users and caregivers. Some technologies are prone to technical failures or errors, resulting in unreliable performance. Particularly, in emergency situations timely detection is critical. Many advanced technologies are inaccessible to certain populations, such as low-income individuals or those living in rural areas with limited access to healthcare resources. The systems that involve constant monitoring or data collection gives raise to privacy concerns, especially among older adults who value their independence and autonomy. Also, acceptance and adoption of new technologies by older adults is challenging, particularly if they perceive them as intrusive or burdensome. Hence, there is a need for a solution which ensures user-friendly design and addresses usability issues which is crucial for widespread adoption.

OBJECTIVES OF THE INVENTION:

[0009] The primary objective of the present invention is to provide a smart footwear for detecting fall of a wearer in an efficient, non-intrusive and user-friendly way.

[0010] Another objective of the present invention is to provide a method for detecting fall of a wearer through improved precision and activity monitoring.

[0011] Yet another objective of the present invention is to provide a method which ensures data security and cost-effective solution for detecting fall which promotes wider acceptance and adoption.


SUMMARY:

[0012] The present invention relates to a smart footwear for detecting fall of a wearer by using IoT and machine learning algorithms. The present invention offers an innovative solution to a critical problem in terms of health and safety.

[0013] According to the present invention, the smart footwear comprises of a microcontroller for processing data, communicating with IoT platform, and triggering alerts, a motion sensor module for detecting motion, orientation, and changes in velocity of wearer, a sound sensor module for detecting sound levels in vicinity of wearer, a Global Positioning System (hereinafter referred GPS) module for enabling geo-fencing capabilities and allowing caregivers to define safe zones for wearer and a Li-ion battery for powering the smart footwear to ensure continuous operation throughout the day.

[0014] In accordance with the present invention, the method for detecting fall of a wearer comprises the steps of collecting data from various sensor modules, preparing the collected data by standardizing the sensor readings, selecting a model to analyze the sensor data and training the model, using the trained model to analyze incoming sensor data and triggering alerts for notifying family members of the wearer and continuously improving accuracy and adaptation using machine learning.

[0015] The present invention introduces a unique approach to fall detection by using a combination of sensor technologies and machine learning algorithms to accurately differentiate between various human activities like jumping, running, sitting, sleeping, standing, walking and falling. The smart footwear is designed to minimize false positives by focusing on patterns of acceleration and angular velocity that distinctly indicate falls, reducing unnecessary alerts or alarms.

Further, the present invention offers the following advantages but not limited to:

• Discrimination between Activities: The present invention has the ability to distinguish between different activities with high accuracy, reducing the likelihood of false alarms
• Machine Learning Integration: The use of machine learning in the present invention refines the detection process, allowing the footwear to adapt to individual patterns of movement and improve over time
• Cost-Effective and Portable Design: The design of the smart footwear focus on portability and cost-effectiveness, making it suitable for real-world applications, particularly in elderly care and health monitoring

These objectives and advantages of the present invention will become more evident from the following detailed description when taken in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS:

The objective of the present invention will now be described in more detail with reference to the accompanying drawing, wherein:

FIG.1 shows the isometric view of the smart footwear disclosed in the present invention for detecting fall;

FIGS. 2A and 2B shows the back view and side view of the smart footwear of the present invention respectively;

FIG. 3 shows the exploded view of Li-ion battery shield used in the smart footwear of the present invention;

FIG. 4 shows the entire circuit system of the smart footwear disclosed in the present invention;

FIG. 5 shows the GSM/GPRS module used in the smart footwear of the present invention;

FIG. 6 shows the Arduino IDE used in the smart footwear of the present invention;

FIGS. 7A and 7B shows the Blynk mobile dashboard and Blynk web dashboard used in the smart footwear of the present invention;

FIG. 8 shows the flow diagram for data pre-processing used in the present invention;

FIG. 9 shows the flow diagram for fall detection algorithm used in the present invention;

FIG. 10 shows the prototype of the smart footwear of the present invention;

FIG. 11 shows the neural network layers used in the present invention;

FIG. 12 shows the scatter plot for determining the threshold for female participants;

FIG. 13 shows the scatter plot for determining the threshold for male participants; and

FIG. 14 shows the Bluetooth terminal app for providing notification.

REFERENCE NUMERALS:

1 - 18650 Li-ion Battery Shield
2 - MPU6050 Motion Sensor
3 - ESP32 Microcontroller
4 - NEO 6M V2 GPS Module
5 - KY-037 Sound Sensor Module

DETAILED DESCRIPTION OF THE INVENTION:

[0016] The present invention relates to a smart footwear for detecting fall of a wearer by using IoT and machine learning algorithms. The present invention offers an innovative solution to a critical problem in terms of health and safety.

[0017] According to the present invention, the smart footwear comprises of a microcontroller for processing data, communicating with IoT platform, and triggering alerts, a motion sensor module for detecting motion, orientation, and changes in velocity of wearer, a sound sensor module for detecting sound levels in vicinity of wearer, a GPS module for enabling geo-fencing capabilities and allowing caregivers to define safe zones for wearer and a Li-ion battery for powering the smart footwear to ensure continuous operation throughout the day.

[0018] In accordance with the present invention, the method for detecting fall of a wearer comprises the steps of collecting data from various sensor modules, preparing the collected data by standardizing the sensor readings, selecting a model to analyze the sensor data and training the model, using the trained model to analyze incoming sensor data and triggering alerts for notifying family members of the wearer and continuously improving accuracy and adaptation using machine learning.

[0019] The detailed description of the components of the smart footwear and the method of working is explained below:

[0020] The smart footwear or smart shoe for detecting fall of a wearer is shown in isometric view in FIG. 1 and comprises of a microcontroller (3), a motion sensor module (2), a sound sensor module (5), a GPS module (4) and a Li-ion battery (1). Also, the back view and side view of the smart footwear are shown in FIGS. 2A and 2B respectively.

Microcontroller:

[0021] The ESP32 microcontroller (3) serves as the brain of the smart shoe. It is responsible for processing data from various sensors, communicating with the IoT platform, and triggering alerts with its Bluetooth capabilities when necessary. The low-power consumption and built-in Wi-Fi capabilities make the microcontroller (3) ideal for IoT applications.

Further, the specifications of the ESP32 microcontroller (3) used in the present invention are listed below:

? Operating Voltage: The ESP32 microcontroller (3) operates within a range of 2.2 V to 3.6 V. This flexibility allows it to work with batteries and other common power supplies.
? Power Requirements: The power consumption of the ESP32 microcontroller (3) varies based on its clock speed, active peripherals, and whether it is in a low-power mode. At full power with maximum clock speed, it consumes around 240 mA. It supports various low-power modes that can significantly reduce power consumption, which is useful for battery-operated or energy-sensitive applications.
? SRAM Capacity: This memory is between 320 KB and 520 KB of SRAM and it is used to store data, variables, and other runtime information required by the microcontroller (3) during program execution. It is volatile as it loses its data when power is removed but it is crucial for running complex programs and handling large datasets efficiently.
? Flash Memory Capacity: The capacity is between 1 MB to 4 MB of built-in Flash memory, which stores program code and non-volatile data. It is slower than SRAM, making it ideal for storing program code, read-only data, and configuration settings. It ensures that programs and settings persist through power cycles.

Given these specifications, the ESP32 microcontroller (3) is widely used in IoT projects, sensor networks, smart devices, and other embedded systems.

Motion sensor module:

[0022] MPU6050 motion sensor module (2) is used in the present invention. The motion sensor module (2) comprises a gyroscope and accelerometer which allows the smart shoe to detect motion, orientation, and changes in velocity. By continuously monitoring these parameters, the smart shoe infers the activities of the wearer and identify sudden movements indicative of a fall.

Sound sensor module:

[0023] KY-037 sound sensor module (5) is used in the present invention. The sound sensor module (5) detects sound levels like analog sound in the vicinity of the shoe / footwear. While primarily used for detecting falls, the footwear also serves other purposes such as detecting loud noises or unusual sounds that may indicate distress. Further, in the present invention, the data detected by sound sensor module (5) is used for analysing its pattern of whether high or low decibels to improve the accuracy.

GPS module:

[0024] The GPS module (4) in the smart shoe enables geo-fencing capabilities, allowing caregivers to define safe zones for the wearer. If the wearer ventures outside these predefined boundaries, the footwear alert caregivers, ensuring they remain within a safe area.

Li-ion battery:

[0025] Powering the smart shoe, the Li-ion battery (1) provides a reliable source of energy to ensure continuous operation throughout the day. The exploded view of 18650 Li-ion Battery shield (1) is shown in FIG. 3.

[0026] The entire circuit system of the smart footwear disclosed in the present invention is shown in FIG. 4.

[0027] In the present invention, two different embodiments are considered for enabling communication between the smart shoe and mobile networks to ensure a reliable way to send fall detection alerts to emergency services and family members.

[0028] In first embodiment, the smart shoe integrates a GSM/GPRS module which is shown in FIG. 5 to connect to mobile networks. This module allows the shoe to send SMS messages or make phone calls to emergency contacts when a fall is detected. However, this embodiment has some drawbacks which are listed below:

? Extra SIM Card: The GSM module requires a dedicated SIM card, which means purchasing and managing additional cellular plans
? 2G Network Dependency: Most low-cost GSM modules like SIM800L operate on 2G networks, which are being phased out in many countries, potentially limiting functionality
? Higher Cost for 4G: While there are GSM modules that work on 4G networks, these are generally more expensive, increasing the overall project cost

[0029] The second embodiment leverages the built-in Wi-Fi and Bluetooth capabilities of the ESP32 microcontroller (3), used as the primary communication method between the smart shoe and a connected mobile device. This embodiment has several advantages over the GSM/GPRS module:

? No Additional SIM Card: This setup does not require an extra SIM card thereby reducing costs and logistical complexity
? Direct Connection to Mobile Devices: The ESP32 (3) is connected directly to a mobile device via Bluetooth or Wi-Fi, enabling instant notifications and seamless communication
? Integration with Mobile Apps: Using Wi-Fi/Bluetooth, the smart shoe sends alerts to a custom mobile app, which can then trigger phone calls to emergency services and family members
? Reduced Hardware Costs: Since the ESP32 (3) already has Wi-Fi and Bluetooth, there is no need to purchase additional hardware modules

[0030] Considering these two embodiments, the second embodiment (ESP32 with Wi-Fi/Bluetooth) provides a more cost-effective, flexible, and future-proof solution. It eliminates the need for additional SIM cards and relies on widely available technology. This reduces the potential for compatibility issues due to network obsolescence and simplifies the communication setup between the smart shoe and the mobile device of wearer.

[0031] The step-by-step method of working of the smart footwear is described below:

Step 1: Data Collection:

[0032] The first step is detecting fall by gathering data from the MPU6050 motion sensor module (3-axis accelerometer and gyroscope) (2) and sound sensor module (5) to capture acceleration, orientation, and sound patterns during various activities. The gathered data is labelled to distinguish between normal activities and fall events by manually annotating the data based on known activities and simulated falls.


Step 2: Data Preparation:

[0033] The second step is to prepare the data. The collected sensor readings are standardized to ensure consistency across different devices and environments. The standardized data is converted into the appropriate format for Convolutional Neural Networks (CNNs) by adding a channel dimension.

Step 3: Model Selection and Training:

[0034] The third step is selecting a model and training. A CNN model with Convolutional layers is used to analyze the sensor data. The Dropout layers are incorporated to reduce overfitting and Batch Normalization for improving training stability. Then, the model is compiled with an optimizer like Adam, a loss function, and evaluation metrics (e.g., accuracy). The model is trained using the prepared data. By implementing early stopping overfitting and checkpointing is avoided to save the best model during training.


Step 4: Data Analyzing and Triggering Alerts:

[0035] The fourth step is to analyze incoming sensor data in real time using the trained CNN model. The model looks for specific patterns indicating a fall, such as rapid changes in acceleration and orientation, combined with abnormal sound patterns. If the model / algorithm detects a fall, it triggers an alert to initiate the notification process.

Step 5: Notification and Communication:

[0036] The fifth step is to send a signal when a fall is detected. This is performed by the integration of a GSM/GPRS module or the built-in Bluetooth of the ESP32 in the smart shoe. The signal is sent to the mobile device of the elderly individual, which then automatically notifies their family members and calls emergency services.

Step 6: Continuous Improvement and Adaptation:

[0037] The final step is implementing machine learning (ML) techniques to improve the accuracy of the algorithm over time. The smart shoe adapts to the unique movement patterns and environmental factors of the wearer. By incorporating a feedback loop, the fall detection algorithm is refined based on new data and real-world use cases.

[0038] In the present invention, Arduino IDE is used for programming Arduino microcontrollers as it is official software. It provides a simple yet powerful code editor tailored for writing and uploading code to Arduino boards. Arduino IDE supports the Arduino programming language, which is based on C/C++. It offers a wide range of libraries and examples to facilitate development. The IDE includes features such as syntax highlighting, code auto-completion, and serial monitor for debugging. Arduino IDE is compatible with various Arduino boards, including the ESP32 microcontroller (3), which is used in the present invention. The programming of Arduino IDE is shown in FIG. 6.

Blynk IOT Platform Integration:

[0039] IoT plays a central role in the present invention by enabling real-time communication between the smart shoe and emergency services, as well as the dashboard for caregivers. Machine learning algorithms are applied to analyze the gathered data, identifying patterns that indicate a fall or unusual activity, and potentially predicting risks before they occur. This integration of IoT and machine learning provides a comprehensive safety solution for seniors, significantly reducing the risk of falls and improving their overall well-being.

[0040] In the present invention, Blynk IoT platform is used. The Blynk is an IoT platform that enables developers to build connected projects with ease. This platform provides a user-friendly interface for real-time data analysis, visualizing and managing data from the smart shoe. A web dashboard is created using Blynk to display real-time sensor data, including sound levels, acceleration, and activity classifications. The caregivers access the dashboard from any internet-enabled device, allowing them to monitor the status of wearer remotely and receive alerts in case of emergencies.

[0041] The Blynk Web Dashboard allows user to create custom dashboards to monitor and control the IoT devices remotely. With Blynk, user can easily interface hardware with the cloud, enabling real-time data visualization and control from anywhere with an internet connection. The platform offers a variety of widgets that user can add to the dashboard, such as buttons, sliders, graphs, and gauges, to interact with user devices. Blynk also provides libraries and APIs for popular hardware platforms, including Arduino, ESP32, and Raspberry Pi, making it easy to integrate with the present invention.

[0042] The Blynk mobile dashboard and Blynk web dashboard used in the smart footwear of the present invention are shown in FIGS. 7A and 7B respectively.

[0043] Further, the data pre-processing is employed in the present invention. The flow diagram of the data pre-processing is shown in FIG. 8. In the data pre-processing stage of Human Activity Recognition (HAR), several crucial steps are undertaken to ensure the quality and suitability of the data for training and evaluation of the machine learning model. First, data cleaning techniques are applied to remove any noisy or irrelevant data points, such as outliers or sensor errors, which negatively impact the performance of the model. Following this, normalization procedures are implemented to scale the sensor data to a common range, facilitating fair comparison and preventing bias towards certain features. Additionally, a sliding window approach is often utilized to segment the continuous sensor data into smaller, overlapping windows, allowing for the extraction of temporal features and capturing the dynamics of human activities over time. Finally, the pre-processed data is split into separate training and testing datasets, typically using a stratified sampling technique to ensure balanced representation of activity classes in both sets. This data split enables the evaluation of the generalization performance of the model on unseen data and helps to prevent overfitting during training. Overall, the data pre-processing stage plays a critical role in preparing the raw sensor data for effective use in training the HAR model, ultimately contributing to the accuracy and reliability of the activity recognition system.

To summarize, the steps for detecting fall of a wearer is shown in the flow diagram in FIG. 9 and is listed below:

1. Read ax, ay, az, rx, ry, and rz from MPU6050 motion sensor (2)
2. Using ML model to train the model to recognize these human activities
3. Getting threshold parameters for activity transitions for a robust system
4. Using those threshold values in Arduino code "if statements"
5. Send notifications to connected mobile phone via BLE 2.0 when fall is detected
6. GPS module (4) reads latitude, longitude, and speed data
7. Sound sensors (5) display analog readings in both web and mobile dashboards

In the present invention, for fall detection and activity monitoring, several parameters play a crucial role in determining the functionality and performance of the smart footwear. They are:

• Voltage Range: 2.2V to 3.6V for the ESP32 microcontroller (3), ensuring compatibility and safe operation within its specified voltage range
• Thresholds for Fall Detection: Threshold values for detecting falls based on acceleration and angular velocity data from the MPU6050 motion sensor module (2) vary depending on factors like body weight, movement dynamics, and individual sensor calibration. These threshold values serve as basic guidelines which help to detect sudden and unexpected changes in movement, indicating a potential fall.

o Stand to Fall: An acceleration spike between 2g to 3g or more on any axis, indicating a sudden drop or impact. An angular velocity spike between 2.618 rad/s to 5.236 rad/s on any axis, indicating sudden rotation during a fall.
o Jump to Fall: An acceleration spike between 3g to 5g on the vertical axis (z-axis), indicating a sudden increase followed by a rapid decrease. An angular velocity spike between 3.491 rad/s to 6.981 rad/s, typically on the vertical axis, indicating a fast spin or tumble.
o Sit to Stand: An increase of about 0.5g to 1g on the vertical axis, indicating a gradual upward movement. A gradual increase in angular velocity, generally between 0.349 rad/s to 0.873 rad/s on any axis, indicating controlled rotational movement.
o Stand to Sit: A decrease of about 0.5g to 1g on the vertical axis, indicating a controlled downward movement. A similar decrease in angular velocity, generally between 20°/s to 50°/s on any axis.
o Run to Fall: An acceleration spike between 3g to 5g, typically on the horizontal axes (x or y), indicating sudden directional changes or impact. An angular velocity spike between 3.491 rad/s to 6.981 rad/s, often across multiple axes, indicating a sudden rotation or twist during a fall.

These threshold values may vary depending on sensor calibration, data processing, individual body dynamics, and context.

• Training Dataset Size: 100 to 1000 labelled samples for training the supervised learning model, ensuring sufficient diversity and representation of activity classes
• Data Split Ratio: 60:40 to 80:20 for training and testing datasets, ensuring an adequate amount of data for model training while still allowing for robust evaluation of performance
• Sampling Rate: 50 Hz to 200 Hz for sensors such as the MPU6050 motion sensor (2) and sound sensor module, balancing between capturing sufficient data for accurate detection and minimizing power consumption
• Sliding Window Size: 1 to 5 seconds for segmenting sensor data into smaller windows, allowing for the extraction of temporal features and capturing the dynamics of human activities over time
• Feature Extraction: 10 to 50 features extracted from sensor data, including time-domain and frequency-domain features such as mean, standard deviation, and Fourier transform coefficients
• Model Hyperparameters: Learning rate (0.001 to 0.01), number of hidden layers (1 to 3), and number of neurons per layer (10 to 100), optimizing the neural network architecture for optimal performance

[0044] A prototype for the present invention has been made but still some necessary modifications have to be done to optimize the design of the smart footwear and to make it visually appealing. The prototype of the smart footwear is shown in FIG. 10 respectively.

[0045] In the present invention, the motion sensor (2) data is collected and trained using a deep learning model, specifically a robust Multi-task Convolutional Neural Network with neural network layers as shown in FIG. 11, to determine threshold values for various fall scenarios. A key feature in the present invention is the approach used to find the threshold values. The model disclosed in the present invention is trained with data from 24 participants (10 female and 14 male) aged between 36 and 70, with an average age of around 56. These 24 subjects performed six different activities (e.g. walking (3), jogging (2), sitting (2), standing (2), going upstairs (3), and going downstairs (3)). The trials were numbered 1-9 (training dataset) and 11-16 (test dataset), with trial 10 excluded. The model uses 12 features, including attitude (roll, pitch, yaw), gravity (x, y, z), user acceleration (x, y, z), and rotation rate (x, y, z). The activity labels categorize movements such as walking, descending stairs (dws), and ascending stairs (ups) into fall scenarios. Additionally, this shoe is trained specifically for gender using datasets labelled by gender (0 for female, 1 for male).

[0046] In the present invention, the input dataset consists of sensor data and labels. A sliding window size of 50 (representing 1 second at a 50Hz sampling rate) and a step size of 10 are used. This method calculates the number of sections based on the sliding window size and step size. The arrays are initialized to store sections of data, activity labels, and gender labels. The data is iterated over to create sections, ensuring consistency of labels within each section. The sections that meet the label consistency criteria are created and stored. Subsequently, a time series is generated for these sections. Training and testing datasets are then prepared for use with a Convolutional Neural Network (CNN) in Keras, a deep learning library. The classes such as Sequential and Model are used to create neural network models, incorporating various layer types including Input, Dense, Flatten, Reshape, Conv2D, MaxPooling2D, and Dropout.

[0047] The model is trained for 20 epochs to simultaneously achieve two primary outputs: gender classification and activity recognition. For activity recognition, a categorical cross-entropy loss function and a Softmax activation function for the output layer are used. For gender classification, a binary cross-entropy loss function and a Sigmoid activation function are applied. The model is compiled with these two loss functions (one for each task), an Adam optimizer, and accuracy as the evaluation metric. After training, the model achieves an accuracy of 0.95111395 for activity recognition and 0.952391 for gender classification. Finally, scatter plots of the testing data are plotted to determine the threshold values. The scatter plot for determining the threshold for female participants is shown in FIG. 12 and scatter plot for determining the threshold for male participants is shown in FIG. 13.

[0048] Also, the Bluetooth terminal app for providing notification is shown in FIG. 14.

[0049] The smart footwear disclosed in the present invention combines multiple advanced sensors to improve the precision of fall detection and activity monitoring. By integrating machine learning algorithms with these sensors, the smart footwear differentiates various human activities and accurately identify falls, reducing false alarms and ensuring reliable performance.

[0050] Unlike other advanced technologies that seem complex or intimidating to certain groups, such as older adults or those with limited technical skills, the smart footwear provides a non-intrusive and user-friendly solution. By embedding the technology within an everyday item like a shoe, it seamlessly integrates into the routine of the wearer, promoting wider acceptance and adoption.

[0051] Further, privacy concerns are another common problem with systems involving continuous monitoring or extensive data collection, especially among older adults who value their independence. The smart footwear addresses these concerns by restricting the data collection to only the parameters necessary for fall detection and activity monitoring. It also ensures data security through encrypted transmission and storage, providing confidentiality and peace of mind for users and their caregivers.

[0052] Also, many existing fall detection and activity monitoring technologies are expensive / require ongoing maintenance and subscription fees, making them inaccessible for some people. The smart footwear solves this by offering a cost-effective solution that uses readily available components, like the ESP32 microcontroller, reducing costs and lowering the barrier to entry.

[0053] To ensure the smart footwear is affordable and accessible to a wide range of users, cost-effective components and materials are utilized without compromising on functionality or reliability. The design optimizations are implemented to minimize production costs while maximizing performance and durability. Bulk purchasing, standardized components, and efficient manufacturing processes contribute to reducing overall production expenses, making the smart footwear economically viable for mass adoption.

[0054] Thus, the present invention encompasses both functionality and aesthetics. The design process involves a careful balance of multiple factors such as comfort, durability, and ease of use. The smart shoe is ergonomically designed to offer comfort and support during daily activities. The present invention also has customization capabilities to adapt to individual users' unique activity patterns thereby enhancing user experience and reliability. Also, in the present invention, there is seamless connectivity with IoT capabilities for real-time notification to family members and emergency services which enables prompt assistance during fall incidents.

[0055] Further, in the present invention, algorithms are carefully developed and optimized to accurately distinguish between normal activities and potential falls, thereby enhancing the safety and well-being of users. Also, the user-friendly web dashboard of the Blynk IoT platform allows caregivers to remotely monitor the status of wearer and receive alerts in case of emergencies, enhancing the overall usability and accessibility of the smart shoe system.

[0056] While the foregoing written description of the invention enables one of ordinary skill to make and use what is considered presently to be the best mode thereof, those of ordinary skill will understand and appreciate the existence of variations, combinations, and equivalents of the specific embodiment, method, and examples herein. The invention should therefore not be limited by the above described embodiment, method, and examples, but by all embodiments and methods within the scope of the invention as claimed.
, Claims:I / WE CLAIM:

1. A smart footwear for detecting fall of a wearer using IoT and machine learning, wherein the smart footwear comprises:
a. a microcontroller (3) for processing data, communicating with IoT platform, and triggering alerts;
b. a motion sensor module (2) for detecting motion, orientation, and changes in velocity of the wearer;
c. a sound sensor module (5) for detecting sound levels in the vicinity of the wearer;
d. a GPS module (4) for enabling geo-fencing capabilities and allowing caregivers to define safe zones for the wearer; and
e. a Li-ion battery (1) for powering the smart footwear to ensure continuous operation.

2. The smart footwear as claimed in claim 1, wherein IoT is Blynk IoT platform with user-friendly web dashboard for real-time data analysis and visualization.

3. The smart footwear as claimed in claim 1, wherein the microcontroller (3) is Arduino ESP32 microcontroller.

4. The smart footwear as claimed in claim 1, wherein the motion sensor module (2) comprises gyroscope and accelerometer.

5. The smart footwear as claimed in claim 1, wherein the sound sensor module (5) detects loud noises, unusual sounds that indicate distress.

6. The smart footwear as claimed in claim 1, wherein the sound levels detected by sound sensor module (5) analyzes its pattern of whether high or low decibels to improve the accuracy.

7. The method for detecting the fall of a wearer using a smart footwear comprises the steps of:
a. collecting data from various sensor modules in the smart footwear;
b. preparing the collected data by standardizing the sensor readings;
c. selecting a model to analyze the sensor data and training the model;
d. using the trained model to analyze incoming sensor data and triggering alerts for notifying family members of the wearer; and
e. improving accuracy and adaptation using machine learning.

8. The method as claimed in claim 7, wherein the trained model uses Convolutional Neural Networks.

Documents