A SYSTEM AND A METHOD FOR HUMAN GESTURE DETECTION IN RADAR IMAGES

Abstract

ABSTRACT A SYSTEM AND A METHOD FOR HUMAN GESTURE DETECTION IN RADAR IMAGES The present disclosure envisages a system (100) and a method (1000) for human gesture detection in radar images. The system (100) comprises a data storage device (102), and a microprocessor (104) communicatively coupled to the data storage device (102) configured to execute one or more processing modules. The receiving module (106) receives low-resolution input images of a plurality of human gestures captured by radar. The first reconstruction module (108) reconstructs the low-resolution input images by implementing the reconstruction technique. The data augmentation module (110) produces an augmented image dataset by generating variations of the images. The second reconstruction module (112) reconstructs the augmented image dataset by implementing a deep learning model. The classification module (114) compares the refined image dataset with the set of classified human gesture image dataset to perform detection of a human gesture. The output module (116) outputs the detected human gesture. FIGURE 1

Specification

Description:FIELD
[0001] The present disclosure relates, in general, to the field of image processing.
[0002] More particularly, embodiments of the present disclosure relate to a system and a method for human gesture detection in radar images.
DEFINITION
[0003] As used in the present disclosure, the following terms are generally intended to have the meaning as set forth below, except to the extent that the context in which they are used to indicate otherwise.
[0004] "Conditional Generative Adversarial Network (CGAN)" refers to a type of Generative Adversarial Network (GAN) where additional information (conditioning variables) is used to control the generation process. This conditioning could be any auxiliary information, such as labels, data classes, or other types of metadata. By conditioning on specific variables, CGANs can generate more targeted and relevant data, making them powerful for applications where control over the generated output is important.
[0005] "Convolutional Neural Network (CNN)" refers to a specialized type of deep neural network commonly used for analyzing visual data. CNNs are especially well-suited for image recognition and classification tasks, where they can automatically and adaptively learn spatial hierarchies in images through layers of convolutions, pooling, and fully connected neurons. CNNs are inspired by the visual processing in animal brains, particularly the way neurons in the visual cortex are organized to respond to overlapping regions of an image.
[0006] "Synthetic Aperture Radar (SAR)" refers to an advanced form of radar used for high-resolution imaging, often from airborne or space borne platforms. Unlike optical imaging systems, SAR is an active remote sensing technology that can capture images in all weather conditions, day or night. SAR operates by emitting radar signals and recording the reflected echoes from the Earth's surface to create detailed images.
BACKGROUND
[0007] The background information herein below relates to the present disclosure but is not necessarily prior art.
[0008] Synthetic Aperture Radar (SAR) images are high-resolution radar-generated images that provide detailed views of the Earth's surface, regardless of weather or lighting conditions. By using active microwave signals, SAR transmits pulses that reflect off surfaces like buildings, forests, and bodies of water, capturing both the intensity and phase of these reflections. SAR images are valuable in numerous fields, including environmental monitoring, military reconnaissance, and disaster management, as they can capture detailed information under clouds, at night, and even in adverse weather. SAR imagery is also used in techniques like Interferometric SAR (InSAR), where images taken at different times are compared to detect surface changes and ground movements with millimetre accuracy.
[0009] Classification of Synthetic Aperture Radar (SAR) images is essential for interpreting and making the most of the detailed information they provide. SAR images contain complex backscatter patterns that vary depending on surface characteristics such as material, texture, moisture, and structure.
[0010] Existing solutions for the classification of Synthetic Aperture Radar (SAR) images leverage a range of approaches, from traditional machine learning techniques to advanced deep learning models.
[0011] However, existing solutions for the classification of Synthetic Aperture Radar (SAR) images face several key challenges that impact their effectiveness. One major issue is the inherent complexity and noise in SAR data, especially speckle noise, which can obscure details and reduce classification accuracy. Traditional machine learning methods, such as Support Vector Machines (SVM) and Random Forests, often require extensive pre-processing and handcrafted feature extraction to achieve acceptable results, which is time-consuming and labour-intensive. Although deep learning models, like Convolutional Neural Networks (CNNs), have shown promise in automating feature extraction, they typically require large amounts of labelled data to generalize effectively. This is problematic since high-quality labelled SAR datasets are limited and costly to produce.
[0012] Therefore, there is a need to develop a system and method for human gesture detection in radar images that alleviate the aforementioned drawbacks.

OBJECTS
[0013] Some of the objects of the present disclosure, which at least one embodiment herein satisfies, are as follows.
[0014] It is an object of the present disclosure to ameliorate one or more problems of the prior art or to at least provide a useful alternative.
[0015] The main object of the present disclosure is to provide a system and a method for human gesture detection in radar images.
[0016] Another object of the present disclosure is to provide a system and a method that utilizes simplified input for Synthetic Aperture Radar (SAR) reconstruction to reduce computational complexity while accurately simulating how radar would detect human-like objects.
[0017] Another object of the present disclosure is to provide a system and a method for human gesture detection in radar images without needing complex or high-resolution inputs.
[0018] Another object of the present disclosure is to provide a system and a method that uses limited training datasets enabling effective performance even with minimal data.
[0019] Another object of the present disclosure is to provide a system and a method that reduces computational costs.
[0020] Another object of the present disclosure is to provide a system and a method that speeds up the SAR reconstruction process.
[0021] Another object of the present disclosure is to provide a system and a method that minimizes the need for extra resources.
[0022] Another object of the present disclosure is to provide a system and a method that reduces the time required for training machine learning models.
[0023] Another object of the present disclosure is to provide a system and a method that minimizes the cost of operation and increases total expected efficiency.
[0024] Another object of the present disclosure is to provide a system and a method that can be tailored for various sectors, including healthcare, autonomous vehicles, and defense, allowing it to meet specific needs across multiple domains.
[0025] Other objects and advantages of the present disclosure will be more apparent from the following description when read in conjunction with the accompanying figures, which are not intended to limit the scope of the present disclosure.
SUMMARY
[0026] This summary is provided to introduce concepts related to the field of image processing. More particularly, embodiments of the invention relate to a system and a method for human gesture detection in radar images. The concepts are further described below in a detailed description. This summary is not intended to identify essential features of the claimed subject matter nor is it intended for use in determining or limiting the scope of the claimed subject matter.
[0027] The present disclosure envisages a system for human gesture detection in radar images. The system comprises a receiving module, a first reconstruction module, a data augmentation module, a second reconstruction module, a classification module, and an output module.
[0028] The receiving module is configured to receive low-resolution input images of a plurality of human gestures captured by radar.
[0029] The first reconstruction module is configured to receive the low-resolution input images from the receiving module. The first reconstruction module is further configured to reconstruct the low-resolution input images by implementing a reconstruction technique for generating a reconstructed image dataset.
[0030] The data augmentation module is configured to receive the reconstructed image dataset from the first reconstruction module. The data augmentation module is further configured to produce an augmented image dataset by implementing a data augmentation technique. The data augmentation module enhances the diversity and robustness of the reconstructed image dataset by generating variations of the images.
[0031] The second reconstruction module is configured to receive the augmented image dataset from the data augmentation module. The second reconstruction module is further configured to reconstruct the augmented image dataset by implementing a deep learning model for generating a refined image dataset including high-quality and realistic radar images.
[0032] The classification module is configured to receive the refined image dataset from the second reconstruction module and a set of classified human gesture image datasets from a data storage device. The classification module is further configured to compare the refined image dataset with the set of classified human gesture image datasets by implementing a pre-trained classification model to perform the detection of a human gesture.
[0033] The output module is configured to receive the detection of a human gesture from the classification module to output the detected human gesture.
[0034] In an embodiment, the low-resolution input images are stickman images.
[0035] In an embodiment, the plurality of human gestures includes sitting gestures, standing gestures, sleeping gestures, falling gestures, and walking gestures.
[0036] In an embodiment, the reconstruction technique is a simulation technique for radar detection of human-like subjects.
[0037] In an embodiment, the reconstructed image dataset includes synthetic aperture radar (SAR)-like radar images.
[0038] In an embodiment, the data augmentation technique ensures the system handles a wider variety of input conditions.
[0039] In an embodiment, the deep learning model is a Conditional Generative Adversarial Network (CGAN).
[0040] In an embodiment, the CGAN generates realistic radar images by learning the conditional mapping between the input images and the corresponding SAR images.
[0041] In an embodiment, the pre-trained classification model includes a training module. The training module is configured to train the system. The training module comprises an input module, an image enhancing module, and a learning module.
[0042] The input module receives the low-resolution input images of a plurality of human gestures captured by radar.
[0043] The image enhancing module enhances the low-resolution input images received from the input module to generate high-quality and realistic radar images which are needed for the training phase.
[0044] The learning module generates the output for training the system by implementing a machine learning model using high-quality and realistic radar images for the detection of a human gesture in a testing module.
[0045] In an embodiment, the pre-trained classification model is a convolutional neural network (CNN) model.
[0046] The present disclosure further envisages a method for human gesture detection in radar images. The said method comprises the steps:
• receiving, by a receiving module, low-resolution input images of a plurality of human gestures captured by a radar;
• reconstructing, by a first reconstruction module, the low-resolution input images received from the receiving module by implementing a reconstruction technique for generating a reconstructed image dataset encompassing a plurality of parameters;
• producing, by a data augmentation module, an augmented image dataset by implementing a data augmentation technique on the reconstructed image dataset received from the reconstruction module to enhance diversity and robustness of the reconstructed image dataset by generating variations of the images;
• reconstructing, by a second reconstruction module, the augmented image dataset received from the data augmentation module by implementing a deep learning model for generating refined image dataset including high-quality and realistic radar images;
• comparing, by a classification module, the refined image dataset received from the second reconstruction module with a set of classified human gesture image datasets received from a data storage device by implementing a pre-trained classification model to perform detection of a human gesture; and
• outputting, by an output module, the detected human gesture received from the classification module.
BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWING
[0047] A system and method for human gesture detection in radar images of the present disclosure will now be described with the help of the accompanying drawings, in which:
[0048] FIGURE 1 illustrates a high-level network architecture of the system for human gesture detection in radar images, in accordance with an embodiment of the present disclosure;
[0049] FIGURE 2 illustrates a block diagram of human gesture classification by integrating Synthetic Aperture Radar (SAR) and Conditional Generative Adversarial Networks (CGAN) approach, in accordance with an embodiment of the present disclosure;
[0050] FIGURE 3 illustrates an architecture of a training module with reference to FIGURE 1;
[0051] FIGURE 4 illustrates an architecture of a generator network to produce SAR-like images, in accordance with an embodiment of the present disclosure;
[0052] FIGURE 5 illustrates an architecture of a discriminator network to distinguish between the original SAR images and CGAN-generated images, in accordance with an embodiment of the present disclosure;
[0053] FIGURE 6 illustrates a five distinct human gestures including walking, standing, falling, sleeping, and sitting, in accordance with an embodiment of the present disclosure;
[0054] FIGURE 7 illustrates a representation of walking gesture losses for CGAN, in accordance with an embodiment of the present disclosure;
[0055] FIGURE 8 illustrates a representation of falling gesture losses for CGAN, in accordance with an embodiment of the present disclosure;
[0056] FIGURE 9 illustrates a representation of the confusion matrix, in accordance with an embodiment of the present disclosure; and
[0057] FIGURES 10A & 10B illustrate a method for human gesture detection in radar images, in accordance with an embodiment of the present disclosure.

LIST OF REFERENCE NUMERALS USED IN THE DESCRIPTION AND DRAWING:
100 System
102 Data storage device
104 Microprocessor
106 Receiving module
108 First reconstruction module
110 Data augmentation module
112 Second reconstruction module
114 Classification module
116 Output module
200 Training module
202 Input module
204 Image enhancing module
206 Learning module

DETAILED DESCRIPTION
[0058] Embodiments, of the present disclosure, will now be described with reference to the accompanying drawing.
[0059] Embodiments are provided so as to thoroughly and fully convey the scope of the present disclosure to the person skilled in the art. Numerous details are set forth, relating to specific components and methods to provide a complete understanding of embodiments of the present disclosure. It will be apparent to the person skilled in the art that the details provided in the embodiments should not be construed to limit the scope of the present disclosure. In some embodiments, well-known apparatus structures, and well-known techniques are not described in detail.
[0060] The terminology used, in the present disclosure, is only for the purpose of explaining a particular embodiment and such terminology shall not be considered to limit the scope of the present disclosure. As used in the present disclosure, the forms "a", "an", and "the" may be intended to include the plural forms as well, unless the context clearly suggests otherwise. The terms, "comprises", "comprising", "including" and "having" are open-ended transitional phrases and therefore, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not forbid the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
[0061] When an element is referred to as being "embodied thereon", "engaged to", "coupled to" or "communicatively coupled to" another element, it may be directly on, engaged, connected, or coupled to the other element. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed elements.
[0062] Synthetic Aperture Radar (SAR) images are valuable in numerous fields, including environmental monitoring, military reconnaissance, and disaster management, as they can capture detailed information under clouds, at night, and even in adverse weather. SAR imagery is also used in techniques like Interferometric SAR (InSAR), where images taken at different times are compared to detect surface changes and ground movements with millimetre accuracy. SAR images are high-resolution radar-generated images that provide detailed views of the Earth's surface, regardless of weather or lighting conditions.
[0063] SAR images contain complex backscatter patterns that vary depending on surface characteristics such as material, texture, moisture, and structure. Classification of Synthetic Aperture Radar (SAR) images is essential for interpreting and making the most of the detailed information they provide.
[0064] Existing solutions for the classification of Synthetic Aperture Radar (SAR) images leverage a range of approaches, from traditional machine learning techniques to advanced deep learning models.
[0065] However, existing solutions for the classification of Synthetic Aperture Radar (SAR) images face several key challenges that impact their effectiveness. One major issue is the inherent complexity and noise in SAR data, especially speckle noise, which can obscure details and reduce classification accuracy. Traditional machine learning methods, such as Support Vector Machines (SVM) and Random Forests, often require extensive pre-processing and handcrafted feature extraction to achieve acceptable results, which is time-consuming and labour-intensive. Although deep learning models, like Convolutional Neural Networks (CNNs), have shown promise in automating feature extraction, they typically require large amounts of labelled data to generalize effectively. This is problematic since high-quality labelled SAR datasets are limited and costly to produce.
[0066] To address the mentioned issues, the present disclosure envisages a system and method for automated tomato leaf disease prediction and remedy recommendation.
[0067] The present invention provides an improvement to existing SAR image reconstruction and classification processes. This invention addresses the significant challenge in machine learning of needing large amounts of labelled data for training. SAR data is difficult to obtain in large quantities, so this system uses synthetic data and machine learning techniques to simulate and enhance radar images.
[0068] By integrating data augmentation and Conditional Generative Adversarial Networks (CGAN), it enhances the quality of SAR images and improves classification accuracy, especially when working with low input data or images. This approach solves the problem of data scarcity and generates high-quality radar images that are more accurate for object or gesture classification, offering a significant advancement over traditional methods.
[0069] The present disclosure envisages a system and method for human gesture detection in radar images. The system 100 for human gesture detection in radar images is described herein with reference to Figures 1 to 9, and a method 1000 for human gesture detection in radar images is described with reference to Figure 10.
[0070] Figure 1 illustrates the high-level network architecture of the system 100 for human gesture detection in radar images, in accordance with an embodiment of the present disclosure. The system 100 includes a data storage device 102 and a microprocessor 104.
[0071] The data storage device 102 is configured to store a set of pre-defined instructions, a set of image processing rules, a reconstruction technique, a pre-trained classification model, and a set of classified human gesture image datasets.
[0072] In an aspect, the set of image processing rules includes image involves manipulating or analyzing images to enhance, transform, or extract meaningful information.
[0073] In an embodiment, the reconstruction technique is a method used to generate or rebuild an image from raw data, often acquired through sensors or other measurement system.
[0074] In an embodiment, the pre-trained classification model is a convolutional neural network (CNN) model.
[0075] In an embodiment, the set of classified human gesture image datasets refers to a collection of organized data used to train and test machine learning models for recognizing human gestures. These datasets typically consist of images or videos labelled with specific gesture categories and are designed to include a diverse range of gestures, lighting conditions, and subjects for robust performance.
[0076] The system 100 may also include a microprocessor 104. The microprocessor 104 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, logic circuitries, and/or any devices that manipulate data based on operational instructions. Among other capabilities, the processor 104 is configured to fetch and execute the one or more pre-determined instructions stored in the data storage device 102.
[0077] The microprocessor 104 may be implemented as a combination of hardware and programming (for example, programmable instructions) to implement one or more functionalities of the processor 104. In the examples described herein, such combinations of hardware and programming may be implemented in several different ways. For example, the programming for the microprocessor 104 may be processor-executable instructions stored on a non-transitory machine-readable storage medium and the hardware for the microprocessor 104 may include a processing resource (for example, one or more processors), to execute such instructions. In the present example, the machine-readable storage medium may store instructions that, when executed by the processing resource, implement the one or more processing modules for the human gesture detection process in radar images. In such examples, the microprocessor 104 may include the machine-readable storage medium storing the instructions and the processing resource to execute the set of predefined instructions for implementing the tomato leaf disease prediction and remedy recommendation process, or the machine-readable storage medium may be separate but accessible to the system 100 and the processing resource. In other examples, the microprocessor 104 may be implemented by electronic circuitry or a printed circuit board.
[0078] The one or more processing modules include a receiving module 106, a first reconstruction module 108, a data augmentation module 110, a second reconstruction module 112, a classification module 114, and an output module 116.
[0079] The receiving module 106 is configured to receive low-resolution input images of a plurality of human gestures captured by radar.
[0080] In an embodiment, the low-resolution input images are stickman images.
[0081] In an embodiment, the plurality of human gestures includes sitting gestures, standing gestures, sleeping gestures, falling gestures, and walking gestures.
[0082] The first reconstruction module 108 is configured to receive the low-resolution input images from the receiving module 106. The first reconstruction module 108 is further configured to reconstruct the low-resolution input images by implementing the reconstruction technique for generating a reconstructed image dataset encompassing a plurality of parameters.
[0083] In an embodiment, the reconstruction technique is a simulation technique for radar detection of human-like subjects.
[0084] The reconstruction module generates simplified stickman images. Simplified stickman images are used as surrogate representations of human-like objects. These inputs are computationally lightweight yet effective in mimicking the radar's interaction with real-world subjects.
[0085] In an embodiment, the reconstructed image dataset includes synthetic aperture radar (SAR)-like radar images.
[0086] The data augmentation module 110 is configured to receive the reconstructed image dataset from the first reconstruction module 108. The data augmentation module 110 is further configured to produce an augmented image dataset by implementing a data augmentation technique to enhance the diversity and robustness of the reconstructed image dataset by generating variations of the images.
[0087] In an embodiment, the data augmentation technique ensures the system handles a wider variety of input conditions.
[0088] The second reconstruction module 108 is configured to receive the augmented image dataset from the data augmentation module 110. The second reconstruction module 108 is further configured to reconstruct the augmented image dataset by implementing a deep learning model for generating a refined image dataset including high-quality and realistic radar images.
[0089] In an embodiment, the plurality of relevant features includes colour histograms, texture features, shape descriptors, and a combination of the features mentioned earlier.
[0090] In an embodiment, the deep learning model is a Conditional Generative Adversarial Network (CGAN).
[0091] In an embodiment, the CGAN generates realistic radar images by learning the conditional mapping between the input images and the corresponding SAR images.
[0092] The classification module 114 is configured to receive the refined image dataset from the second reconstruction module 112 and a set of classified human gesture image datasets from a data storage device 102. The classification module 114 is further configured to compare the refined image dataset with the set of classified human gesture image datasets by implementing a pre-trained classification model to perform detection of a human gesture.
[0093] In an embodiment, the pre-trained classification model is a convolutional neural network (CNN) model.
[0094] In an aspect, the pre-trained classification model includes a training module 200.
[0095] The output module 116 is configured to receive the detection of a human gesture from the classification module 114 to output the detected human gesture.
[0096] FIGURE 2 illustrates a block diagram of the process for human gesture classification by integrating Synthetic Aperture Radar (SAR) and Conditional Generative Adversarial Networks (CGAN), in accordance with an embodiment of the present disclosure.
[0097] The present disclosure introduces an innovative approach for Synthetic Aperture Radar (SAR) image reconstruction and analysis using simplified stickman images as input, significantly reducing computational complexity while accurately simulating radar detection of human-like objects. By employing data augmentation techniques, the system enhances SAR datasets to improve generalization and performance, even with limited data. Conditional Generative Adversarial Networks (CGANs) refine these augmented images, producing high-quality outputs that closely resemble realistic radar captures. This integration of SAR reconstruction, CGAN refinement, and object classification ensures accurate identification of objects or human gestures, such as sitting or walking.
[0098] The steps in the process for human gesture classification by integrating Synthetic Aperture Radar (SAR) and Conditional Generative Adversarial Networks (CGAN) include:
1. Input (Stickman Images): The system begins by using simplified or low-resolution images, such as stickman images, which likely serve as a form of simulated input for training and testing purposes.
2. SAR Image Reconstruction: The first step involves reconstructing the stickman images into SAR-like radar images, simulating how radar would detect the reflected signals from simple objects.
3. Data Augmentation: To improve the diversity and robustness of the dataset, the reconstructed SAR images are subjected to data augmentation. This step generates variations of the images, ensuring that the system can handle a wider variety of input conditions. Even though data augmentation can produce repetitive patterns and lack complex variations, limiting the model's ability to generalize. To overcome that CGANs address these issues by generating more realistic and diverse outputs.
4. Reconstruction Using CGAN: The augmented SAR images are further refined using a Conditional Generative Adversarial Network (CGAN). This Deep-learning model generates more realistic radar images by learning the conditional mapping between the input (stickman images) and the corresponding SAR images. The CGAN helps in generating good-quality SAR images that closely resemble real radar captures.
5. Classification: Finally, the output from the CGAN is passed through a classification algorithm CNN, where the system identifies and classifies the human gesture represented in the radar images.
[0099] FIGURE 3 illustrates an architecture of a training module with reference to FIGURE 1.
[00100] The training module 200 is configured to train the system 100. The training module 200 comprises an input module 202, an image enhancing module 204, and a learning module 206.
[00101] The input module 202 receives the low-resolution input images of a plurality of human gestures captured by radar.
[00102] The image enhancing module 204 enhances the low-resolution input images received from the input module 202 to generate high-quality and realistic radar images which are needed for the training phase.
[00103] The learning module 206 generates the output for training the system 100 by implementing machine learning model using high-quality and realistic radar images for the detection of a human gesture in a testing module.
[00104] In an aspect, the training module 200 is a convolutional neural network (CNN) model.
[00105] FIGURE 4 illustrates an architecture of a generator network to produce SAR-like images, in accordance with an embodiment of the present disclosure;
[00106] The system 100 is trained to produce SAR-like images conditioned on the stickman images using a generator network as shown in FIGURE 4.
[00107] FIGURE 5 illustrates an architecture of a discriminator network to distinguish between the original SAR images and CGAN-generated images, in accordance with an embodiment of the present disclosure.
[00108] The system 100 is trained to distinguish between the original SAR images and CGAN-generated images using a discriminator network as shown in FIGURE 5
[00109] FIGURES 10A & 10B illustrate a method 1000 for human gesture detection in radar images, in accordance with an embodiment of the present disclosure. The order in which the method 1000 is described is not intended to be construed as a limitation, and any number of the described method steps can be combined in any appropriate order to carry out the method 1000 or an alternative method. Additionally, individual steps may be deleted from the method 1000 without departing from the scope of the subject matter described herein. The method 1000 for human gesture detection in radar images is executed by the system 100. The method 1000 includes the following steps:
[00110] In method step 1002, the method 1000 comprises receiving, by a receiving module 106, low-resolution input images of a plurality of human gestures captured by a radar.
[00111] In method step 1004, the method 1000 comprises reconstructing, by a first reconstruction module 108, the low-resolution input images received from the receiving module 106 by implementing a reconstruction technique for generating a reconstructed image dataset encompassing a plurality of parameters.
[00112] In method step 1006, the method 1000 comprises producing, by a data augmentation module 110, an augmented image dataset by implementing a data augmentation technique on the reconstructed image dataset received from the first reconstruction module 108 to enhance the diversity and robustness of the reconstructed image dataset by generating variations of the images.
[00113] In method step 1008, the method 1000 comprises reconstructing, by a second reconstruction module 112, the augmented image dataset received from the data augmentation module by implementing a deep learning model for generating refined image dataset including high-quality and realistic radar images.
[00114] In method step 1010, the method 1000 comprises comparing, by a classification module 114, the refined image dataset received from the second reconstruction module 112 with a set of classified human gesture image dataset received from a data storage device 102 by implementing a pre-trained classification model to perform detection of a human gesture.
[00115] In method step 1012, the method 1000 comprises outputting, by an output module 116, the detected human gesture received from the classification module 114.
WORKING EXAMPLE
[00116] A working example of the present disclosure is as following:
[00117] The steps in the process include:
1. Input Data: Stickman Gesture Images
Gestures: The system 100 involves five distinct human gestures that are sitting, standing, sleeping, falling, and walking.
Initial Dataset: Each gesture has approximately 10 to 12 simplified stickman images, bringing the total to around 60 images in the initial dataset. The human gestures are shown in FIGURE 6.
2. SAR Image Reconstruction:
Process: Each stickman image is reconstructed into a SAR image using MATLAB. This process generates radar-like representations of the stickman gestures, simulating how radar systems would detect such movements.
Output: For each stickman image, a corresponding SAR image is produced, resulting in around 60 SAR images.


3. Data Augmentation:
Augmentation Techniques: Each SAR image is augmented 20 times using techniques like rotation, scaling, flipping, translation, and brightness adjustment.
Dataset Size: The data augmentation process disclosed in the working example increases the total number of images to approximately 1,200 (60 SAR images * 20 augmentations per image).
4. CGAN for SAR Image Refinement:
Network Architecture: A CGAN is implemented using Deep learning to enhance the quality of the SAR-reconstructed images.
Refinement: 1200 augmented SAR images are fed into a Conditional Generative Adversarial Network (CGAN) for further refinement.
5. Training Procedure:
Generator: Trained to produce SAR-like images conditioned on the stickman images using a generator network as shown in FIGURE 4.
Discriminator: Trained to distinguish between the original SAR images and CGAN-generated images using a discriminator network as shown in FIGURE 5.
Epochs: The CGAN is trained over 500 epochs for each image, with real-time evaluation of image similarity and quality.
Loss Function: Mean Squared Error (MSE) is used to calculate the accuracy between the generated images and the target outputs.
After both augmentation and CGAN processing, the final dataset comprises 5000 augmented and CGAN-refined SAR images for gesture classification.
Performance Graph:
The performance graphs for walking gestures and falling gestures are shown in FIGURES 7 and 8.
6. Classification Using CNN( EfficientNet ):
Network Architecture: EfficientNet is actually based on CNNs, but it introduces a unique scaling approach to optimize CNN architectures in a way that balances performance with efficiency.
Since the input dataset is smaller and simpler, EfficientNet's efficient parameterization helps avoid over fitting. It achieves better accuracy without needing a very large dataset, which is often required by larger CNN models. EfficientNet models allow us to utilize transfer learning effectively, enhancing our results even with a limited amount of training data.
[00118] Training: The system is trained on 80% of the dataset and validated on 20%. Data augmentation ensured that the model is trained on a variety of gestures and viewing angles.
[00119] Evaluation Metric: The confusion matrix is computed to evaluate classification performance after training of 25 epochs on the basis of accuracy as shown in FIGURE 9.
[00120] Final Results and Observations:
• The combination of SAR image reconstruction, data augmentation, and CGAN refinement produced high-quality, radar-like images suitable for gesture classification.
• The classification yielded promising results, with significant accuracy in gesture detection from SAR images.
• Data augmentation and CGAN were particularly effective in improving image clarity and classification accuracy with limited datasets.
[00121] The present disclosure disclosed an efficient system for generating, reconstructing, and classifying SAR images for human activity detection, using a combination of data augmentation and CGAN. This approach includes the ability to generate high-quality SAR images from minimal data. This disclosure significantly improves the accuracy and practicality of radar-based human activity classification and object detection, making it more accessible for real-world applications.
[00122] The foregoing description of the embodiments has been provided for purposes of illustration and is not intended to limit the scope of the present disclosure. Individual components of a particular embodiment are generally not limited to that particular embodiment but, are interchangeable. Such variations are not to be regarded as a departure from the present disclosure, and all such modifications are considered to be within the scope of the present disclosure.
APPLICATIONS
[00123] The system 100 and method 1000 for human gesture detection in radar images using a system 100 have a wide range of uses, applications, and benefits in the field of image processing. Some of the major applications and potential uses for the present invention include:
1. Surveillance and Security: The system 100 can be applied in radar-based surveillance systems for detecting human activities or objects, making it highly useful for military, border security, and disaster response operations. The ability to process low-resolution images and reconstruct them into radar data allows for detection and classification, even in limited data conditions.
2. Remote Sensing and Environmental Monitoring: The SAR image reconstruction process of the system 100 can be used for monitoring environmental changes such as deforestation, natural disasters, or agricultural conditions. It offers precise classification of radar data to track and categorize landscape changes or movement of objects in a region.
3. Autonomous Vehicles: It can be used for object recognition and classification in autonomous driving, improving safety and navigation in low-visibility conditions.
4. Healthcare Monitoring: It can assist in non-invasive patient monitoring, recognizing postures and movements in medical environments for fall detection or patient care.
5. Military and Defence: In defense applications, the system 100 can be used for reconnaissance, identifying targets and movements in difficult-to-access areas using radar imaging
TECHNICAL ADVANCEMENTS AND ECONOMIC SIGNIFICANCE
[00124] The present disclosure described herein above has several technical advantages including, but not limited to, a system and a method for human gesture detection in radar images, which:
• provides the user(s) with a system and a method that utilizes simplified input for Synthetic Aperture Radar (SAR) reconstruction to reduce computational complexity while accurately simulating how radar would detect human-like objects;
• provides the user(s) with a system and a method for human gesture detection in radar images without needing complex or high-resolution inputs;
• provides the user(s) with a system and a method that uses limited training datasets enabling effective performance even with minimal data; and
• provides the user(s) with a system and a method that reduces the time required for training machine learning models.
[00125] The present disclosure described herein above has several economic advantages including, but not limited to:
• cost-effectiveness, compared to the state-of-the-art equipment used for human gesture detection in radar images based on sophisticated technologies as it can minimize the cost of operation;
• flexibility in real-world applications, can be tailored for various sectors, including healthcare, autonomous vehicles, and defence, allowing it to meet specific needs across multiple domains;
• time-saving, compared to the state-of-the-art equipment used for human gesture detection in radar images based on sophisticated technologies as it speeds up the SAR reconstruction process
• reduced complexity, compared to the state-of-the-art equipment used for human gesture detection in radar images based on sophisticated technologies as it minimizes the need for extra resources; and
• accuracy, compared to the state-of-the-art equipment used for human gesture detection in radar images based on sophisticated technologies as it can increase total expected efficiency.
[00126] The embodiments herein and the various features and advantageous details thereof are explained with reference to the non-limiting embodiments in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein may be practiced and to further enable those of skill in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.
[00127] The foregoing description of the specific embodiments so fully reveals the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and/or adapt for various applications such specific embodiments without departing from the generic concept, and, therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of preferred embodiments, those skilled in the art will recognize that the embodiments herein can be practiced with modification within the spirit and scope of the embodiments as described herein.
[00128] The use of the expression "at least" or "at least one" suggests the use of one or more elements or ingredients or quantities, as the use may be in the embodiment of the disclosure to achieve one or more of the desired objects or results.
[00129] Any discussion of documents, acts, materials, devices, articles, or the like that has been included in this specification is solely for the purpose of providing a context for the disclosure. It is not to be taken as an admission that any or all of these matters form a part of the prior art base or were common general knowledge in the field relevant to the disclosure as it existed anywhere before the priority date of this application.
[00130] The numerical values mentioned for the various physical parameters, dimensions, or quantities are only approximations and it is envisaged that the values higher/lower than the numerical values assigned to the parameters, dimensions or quantities fall within the scope of the disclosure, unless there is a statement in the specification specific to the contrary.
[00131] While considerable emphasis has been placed herein on the components and component parts of the preferred embodiments, it will be appreciated that many embodiments can be made and that many changes can be made in the preferred embodiments without departing from the principles of the disclosure. These and other changes in the preferred embodiment as well as other embodiments of the disclosure will be apparent to those skilled in the art from the disclosure herein, whereby it is to be distinctly understood that the foregoing descriptive matter is to be interpreted merely as illustrative of the disclosure and not as a limitation.
, Claims:WE CLAIM:
1. A system (100) for human gesture detection in radar images, said system (100) comprising:
a receiving module (106) configured to receive low-resolution input images of a plurality of human gestures captured by radar;
a first reconstruction module (108) configured to receive the low-resolution input images from the receiving module (106), to reconstruct the low-resolution input images by implementing the reconstruction technique for generating a reconstructed image dataset encompassing a plurality of parameters;
a data augmentation module (110) configured to receive the reconstructed image dataset from the first reconstruction module (108) producing an augmented image dataset by implementing a data augmentation technique to enhance the diversity and robustness of the reconstructed image dataset by generating variations of the images;
a second reconstruction module (112) configured to receive the augmented image dataset from the data augmentation module (110) to reconstruct the augmented image dataset by implementing a deep learning model for generating refined image dataset including high-quality and realistic radar images;
a classification module (114) configured to receive the refined image dataset from the second reconstruction module (112) and a set of classified human gesture image datasets from a data storage device (102), and further configured to compare the refined image dataset with the set of classified human gesture image dataset by implementing a pre-trained classification model to perform detection of a human gesture; and
an output module (116) configured to receive the detection of a human gesture from the classification module (114) to output the detected human gesture.
2. The system (100) as claimed in claim 1, wherein the low-resolution input images are stickman images.
3. The system (100) as claimed in claim 1, wherein the plurality of human gestures includes sitting gestures, standing gestures, sleeping gestures, falling gestures, and walking gestures.
4. The system (100) as claimed in claim 1, wherein the reconstruction technique is a simulation technique for radar detection of human-like subjects.
5. The system as claimed in claim 1, wherein the reconstructed image dataset includes synthetic aperture radar (SAR)-like radar images.
6. The system (100) as claimed in claim 1, wherein the data augmentation technique ensures the system to handle a wider variety of input conditions.
7. The system (100) as claimed in claim 1, wherein the deep learning model is a Conditional Generative Adversarial Network (CGAN).
8. The system (100) as claimed in claim 1, wherein the CGAN generates realistic radar images by learning the conditional mapping between the input images and the corresponding SAR images.
9. The system (100) as claimed in claim 1, wherein the pre-trained classification model includes a training module (200) configured to train the system (100), the training module (200) comprising:
an input module (202) to receive low-resolution input images of a plurality of human gestures captured by radar;
an image enhancing module (204) to enhance the low-resolution input images received from the input module (202) to generate high-quality and realistic radar images which are needed for the training phase; and
a learning module (206) to generate the output for training the system (100) by implementing a machine learning model using high-quality and realistic radar images for the detection of a human gesture in a testing module.
10. The system (100) as claimed in claim 1, wherein the pre-trained classification model is a convolutional neural network (CNN) model.
11. A method (1000) for human gesture detection in radar images, said method (1000) comprising:
receiving (1002), by a receiving module (106), low-resolution input images of a plurality of human gestures captured by a radar;
reconstructing (1004), by a first reconstruction module (108), the low-resolution input images received from the receiving module (106) by implementing a reconstruction technique for generating a reconstructed image dataset encompassing a plurality of parameters;
producing (1006), by a data augmentation module (110), an augmented image dataset by implementing a data augmentation technique on the reconstructed image dataset received from the first reconstruction module (108) to enhance diversity and robustness of the reconstructed image dataset by generating variations of the images;
reconstructing (1008), by a second reconstruction module (112), the augmented image dataset received from the data augmentation module () by implementing a deep learning model for generating refined image dataset including high-quality and realistic radar images;
comparing (1010), by a classification module (114), the refined image dataset received from the second reconstruction module (112) with a set of classified human gesture image datasets received from a data storage device (102) by implementing a pre-trained classification model to perform detection of a human gesture; and
outputting (1012), by an output module (116), the detected human gesture received from the classification module (114).
Dated this 22nd day of November, 2024

_______________________________
MOHAN RAJKUMAR DEWAN, IN/PA - 25
of R.K.DEWAN & CO.
Authorized Agent of Applicant

TO,
THE CONTROLLER OF PATENTS
THE PATENT OFFICE, AT CHENNAI

Documents