Gait Identification Research Articles

Piloting a Novel Computational Framework for Identifying Prosthesis-Specific Contributions to Gait Deviations.

This paper introduces a novel computational framework for evaluating above-knee prostheses, addressing a major challenge in gait deviation studies: distinguishing between prosthesis-specific and patient-specific contributions to gait deviations. This innovative approach utilizes three separate computational models to quantify the changes in gait dynamics necessary to achieve a set of ideal gait kinematics across different prosthesis designs. The pilot study presented here employs a simple two-dimensional swing-phase model to conceptually demonstrate how the outcomes of this three-model framework can assess the extent to which prosthesis design impacts a user's ability to replicate the dynamics of able-bodied gait. Furthermore, this framework offers potential for optimizing passive prosthetic devices for individual patients, thereby reducing the need for real-life experiments, clinic visits, and overcoming rehabilitation challenges.

Evaluating CNN Models for Gait Recognition: A Study on the CASIA-B Dataset

Gait recognition, a form of biometric identification, has garnered considerable interest for its ability to identify individuals from a distance. Yet, its effectiveness depends significantly on the precision of the underlying categorization models. This study thoroughly evaluates many Convolutional Neural Network (CNN) models to determine their effectiveness in properly identifying and categorizing gait patterns. We train and test several CNN architectures using advanced deep learning techniques on the widely known CASIA-B dataset, which is a benchmark dataset for gait recognition research. The analyses involve models like as VGG16, VGG19, NASNetLarge, NASNetMobile, EfficientNetB0, and Xception, each well-known for their effectiveness in picture categorization tasks. We evaluate the performance of these models by conducting thorough experiments and detailed analysis, focusing on accuracy, validation loss, and other pertinent metrics. The Xception model demonstrated the highest accuracy of 97.17% of the models assessed. This model frequently surpasses similar models, demonstrating its strength and effectiveness in precisely recognizing gait patterns. On the other hand, the accuracy of other models varied from 9.78% to 93.85%, demonstrating the variable levels of efficacy among various designs. The findings have significant implications for the creation and implementation of gait recognition systems. The high precision demonstrated by the Xception model highlights the promise of CNN-based methods in progressing gait identification technology. Our work highlights the significance of choosing suitable model architectures for achieving the best performance in gait detection tasks. In the future, research may focus on using larger datasets and exploring other CNN architectures to improve the accuracy and reliability of gait detection systems. Our goal is to enhance gait recognition technology to improve biometric identification for security, surveillance, and healthcare applications. GUB JOURNAL OF SCIENCE AND ENGINEERING, Vol 10(1), 2023 P 17-26

Emerging trends in gait recognition based on deep learning: a survey.

Gait recognition, a biometric identification method, has garnered significant attention due to its unique attributes, including non-invasiveness, long-distance capture, and resistance to impersonation. Gait recognition has undergone a revolution driven by the remarkable capacity of deep learning to extract complicated features from data. An overview of the current developments in deep learning-based gait identification methods is provided in this work. We explore and analyze the development of gait recognition and highlight its uses in forensics, security, and criminal investigations. The article delves into the challenges associated with gait recognition, such as variations in walking conditions, viewing angles, and clothing as well. We discuss about the effectiveness of deep neural networks in addressing these challenges by providing a comprehensive analysis of state-of-the-art architectures, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and attention mechanisms. Diverse neural network-based gait recognition models, such as Gate Controlled and Shared Attention ICDNet (GA-ICDNet), Multi-Scale Temporal Feature Extractor (MSTFE), GaitNet, and various CNN-based approaches, demonstrate impressive accuracy across different walking conditions, showcasing the effectiveness of these models in capturing unique gait patterns. GaitNet achieved an exceptional identification accuracy of 99.7%, whereas GA-ICDNet showed high precision with an equal error rate of 0.67% in verification tasks. GaitGraph (ResGCN+2D CNN) achieved rank-1 accuracies ranging from 66.3% to 87.7%, whereas a Fully Connected Network with Koopman Operator achieved an average rank-1 accuracy of 74.7% for OU-MVLP across various conditions. However, GCPFP (GCN with Graph Convolution-Based Part Feature Polling) utilizing graph convolutional network (GCN) and GaitSet achieves the lowest average rank-1 accuracy of 62.4% for CASIA-B, while MFINet (Multiple Factor Inference Network) exhibits the lowest accuracy range of 11.72% to 19.32% under clothing variation conditions on CASIA-B. In addition to an across-the-board analysis of recent breakthroughs in gait recognition, the scope for potential future research direction is also assessed.

Gait-Guard: Turn-aware Freezing of Gait Detection for Non-intrusive Intervention Systems.

Freezing of gait significantly reduces the quality of life for Parkinson's disease patients by increasing the risk of injurious falls and reducing mobility. Real-time intervention mechanisms promise relief from these symptoms, but require accurate real-time, portable freezing of gait detection systems to be effective. Current real-time detection systems have unacceptable false positive freezing of gait identification rates to be adopted by the patients for real-world use. To rectify this, we propose Gait-Guard, a closed-loop, real-time, and portable freezing of gait detection and intervention system that treats symptoms in real-time with a low false positive rate. We collected 1591 freezing of gait events across 26 patients to evaluate Gait-Guard. Gait-Guard achieved a 112% reduction in the false positive intervention rate when compared with other validated real-time freezing of gait detection systems, and detected 96.5% of the true positives with an average intervention latency of just 378.5ms in a subject-independent study, making Gait-Guard a practical system for patients to use in their daily lives.

Gait-based identification using wearable multimodal sensing and attention neural networks

Advances in sensor technology have sparked a wave of innovation and progress in the domain of wearable devices. However, ensuring the security of these devices remains a critical concern for their users. This study proposes a novel identification framework that employs custom multimodal sensing insoles and deep learning techniques to contribute to the data security of wearable devices in a user-transparent manner. The insole incorporates an inertial sensing unit and ten force sensors to capture kinematic and kinetic information. For this implementation, we propose a lightweight multiclass classification deep model based on depthwise separable convolution and an enhanced attention mechanism, which operates over multiple signal channels. In designing the recognition model, we focused on two pivotal factors: (1) achieving an optimal balance between computational complexity and recognition accuracy, which is essential for models that operate on devices with limited computational power, and (2) introducing an enhanced attention mechanism that ensures high recognition accuracy without the stringent requirement for precise temporal alignment between sensor channels, which can be technically challenging to attain. We conducted extensive experiments to validate the proposed approach, utilizing a walking dataset collected from 82 young subjects. The results indicate that when using data from the multimodal sensing insole, a high recognition accuracy of 97.96 % for person identification is achieved, with a notably low model parameter count of 22,462. The results support the feasibility and effectiveness of sensing footwear-based gait identification. The proposed approach holds significant potential for enhancing user authentication methods across a broad range of scenarios while ensuring transparency and user-friendliness.

Passive Multiuser Gait Identification Through Micro-Doppler Calibration Using mmWave Radar

Passive Multiuser Gait Identification Through Micro-Doppler Calibration Using mmWave Radar

HEALTH PROMOTION BY TEACHING DUAL TASK EXERCISE AND EMPOWERING CADRES ON GAIT IMPROVEMENTAS WELL AS THE BALANCE OF THE ELDERLY IN GOWA REGENCY

Walking and balance disorders are the most common causes of falls in the elderly and often lead to injury and disability, Gait and balance disorders are usually multifactorial and require a comprehensive examination to determine contributing factors and appropriate interventions. Most of the gait changes that occur in the elderly are related to underlying medical conditions, walking and balance disorders should not be viewed simply as inevitable consequences of aging. Early identification of gait and balance disorders and determination of appropriate interventions can prevent dysfunction and loss of independence. This community service is carried out in Gowa Regency, South Sulawesi Province. The intervention given to the elderly is to provide and teach how to carry out dual task exercise interventions in postural control, carry out health promotion and empowerment of cadres are given in the right way and carried out continuously to improve the gait and balance of the elderly in Gowa Regency

CDBMGCIG: Design of a Cross-Domain Bioinspired Model for identification of Gait Components via Iterated GANs

This Gait identification assists in recognition of human body components from temporal image sequences. Such components consist of connected-body entities including head, upper body, lower body regions. Existing Gait recognition models use deep learning methods including variants of Convolutional Neural Networks (CNNs), Q-Learning, etc. But these methods are either highly complex, or do not perform well under complex background conditions. Moreover, most of these models are validated on a specific environmental condition, and cannot be scaled for general-purpose deployments. To overcome these issues, this text proposes design of a novel cross-domain bioinspired model for identification of gait components via Iterated Generative Adversarial Networks (IGANs). The proposed model initially extracts multidomain pixel-level feature sets from different images. These include frequency components via Fourier analysis, entropy components via Cosine analysis, spatial components via Gabor analysis, and window-based components via Wavelet &Convolutional analysis. These feature sets are processed via a Grey Wolf Optimization (GWO) Model, which assists in identification of high-density & highly variant features for different gait components. These features are classified via an iterated GAN, which comprises of Generator & Discriminator ssModels that assist in evaluating connected body components. These operations generate component-level scores that assist in identification of gait from complex background images. Due to which, the proposed model was observed to achieve 9.5% higher accuracy, 3.4% higher precision, and 2.9% higher recall than existing gait identification methods. The model also uses iterative learning, due to which its accuracy is incrementally improved w.r.t. number of evaluated image sets.

A Control Strategy for Pneumatically Powered Below-Hip Orthosis

ObjectivesThe role of controllers is inevitable in the design of powered orthosis to achieve ideal gait characteristics. Despite the fact that electrical actuators are preferred for most of the orthoses, a pneumatic actuator proves to have low cost and less weight. In this study, suitable controllers are designed and implemented for the knee and hip joints of a pneumatically actuated orthosis for the afflicted people. Material and MethodsDifferent controllers (P, PI and PID) were tested for the position control of the orthosis by proper tuning of gain constants (KP, KI, KD). By using Lagrange Euler Method, the optimal trajectory for the knee and hip joints were determined for the pneumatic system. Particle swarm optimization (PSO) based PID controller was further employed for optimizing the gain constants. ResultsWith the healthy gait as reference, the knee and hip reference angles were manually set in the PID controller. The subject was made to walk five times at a distance of 5 m and the average knee and hip angles were calculated based on the gait trials. Knee and hip angles varied 0 to 45° and 0 to 35° for healthy subjects while they varied 0 to 41° and 0 to 45° for the implemented pneumatic leg. The values of gain constants obtained in manual tuning matched with the PSO based controller at 25th iteration and the best fitness function was chosen with least error (0.7011). ConclusionThe prototype of the orthosis is fabricated and the response of PID controller was found to be acceptable for a desired pressure (5 bar) with an angular velocity of 3 deg/s. Using a PID controlled pneumatic orthosis, exhibited less oscillation and showed an improved steady-state error when compared to the other controllers, thereby replicating healthy gait. A global best position with minimum error was obtained using PSO to find optimal controller gain constants.

GAIT Technology for Human Recognition using CNN

Gait is a distinctive biometric characteristic that can be detected from a distance; as a result, it has several uses in social security, forensic identification, and crime prevention. Existing gait identification techniques use a gait template, which makes it difficult to keep temporal information, or a gait sequence, which maintains pointless sequential limitations and loses the ability to portray a gait. Our technique, which is based on this deep set viewpoint, is immune to frame permutations and can seamlessly combine frames from many videos that were taken in various contexts, such as diversified watching, angles, various outfits, or various situations for transporting something. According to experiments, our single-model strategy obtains an average rank-1 accuracy of 96.1% on the CASIA-B gait dataset and an accuracy of 87.9% on the OU-MVLP gait dataset when used under typical walking conditions. Our model also demonstrates a great degree of robustness under numerous challenging circumstances. When carrying bags and wearing a coat while walking, it obtains accuracy on the CASIA-B of 90.8% and 70.3%, respectively, greatly surpassing the best approach currently in use. Additionally, the suggested method achieves a satisfactory level of accuracy even when there are few frames available in the test samples; for instance, it achieves 85.0% on the CASIA-B even with only 7 frames.

A diverse and multi-modal gait dataset of indoor and outdoor walks acquired using multiple cameras and sensors

Gait datasets are often limited by a lack of diversity in terms of the participants, appearance, viewing angle, environments, annotations, and availability. We present a primary gait dataset comprising 1,560 annotated casual walks from 64 participants, in both indoor and outdoor real-world environments. We used two digital cameras and a wearable digital goniometer to capture visual as well as motion signal gait-data respectively. Traditional methods of gait identification are often affected by the viewing angle and appearance of the participant therefore, this dataset mainly considers the diversity in various aspects (e.g., participants’ attributes, background variations, and view angles). The dataset is captured from 8 viewing angles in 45° increments along-with alternative appearances for each participant, for example, via a change of clothing. The dataset provides 3,120 videos, containing approximately 748,800 image frames with detailed annotations including approximately 56,160,000 bodily keypoint annotations, identifying 75 keypoints per video frame, and approximately 1,026,480 motion data points captured from a digital goniometer for three limb segments (thigh, upper arm, and head).

Bespoke Fuzzy Logic Design to Automate a Better Understanding of Running Gait Analysis.

Running gait assessment and running shoe recommendation is important for the injury prevention of runners who exhibit different skill-levels and running styles. Traditionally, running gait assessment for shoe recommendation relies upon a combination of trained professionals (e.g., sports-therapists, physiotherapists) and complex equipment such as motion or pressure sensors, often incurring a high-cost to the consumer. Despite this, assessments are still prone to subjectivity, and may differ between assessors. Alternatively, methods to provide low-cost, reproduceable gait assessment has become a necessity, especially within a habitual (low-resource) context, with many traditional methods generally unavailable due to the need of professional assistance and more recently the COVID-19 pandemic. Fuzzy logic has shown to be an effective tool in the assessment and identification of gait by providing the potential for a high-accuracy methodology, while retaining a low computational cost; ideal for applications within embedded systems. Here, we present a novel shoe recommendation fuzzy inference system from the classification of two key running gait parameters, foot strike and pronation from a single foot mounted internet of thing (IoT) enabled wearable inertial measurement unit. The fuzzy approach provides excellent (ICC > 0.9) accuracy, while significantly increasing the resolution of the gait assessment technique, providing a more detailed running gait analysis.

Privacy-Aware Gait Identification With Ultralow-Dimensional Data Using a Distance Sensor

As one of the most natural user behaviors, walking has been widely focused on developing personal identification systems due to its unique biometric authentication features. Popular visual solutions are usually affected by various environmental conditions, and their redundant user information (e.g., body type and appearance) makes it more challenging for users to maintain privacy and security. This paper proposes a distance sensor–based gait identification system that uses only one-dimensional data with a simple system structure. Specifically, a time-of-flight sensor was placed in front of a walking person, and a time series of distances was acquired. We extracted gait features from the data by calculating the velocity and acceleration curves and identifying individuals using a random forest classifier. We evaluated our system on 10 users using leave-one-out cross-validation. The average identification accuracy was 91.05% for 10 users. This study shows that gait recognition is possible using only one-dimensional time-series data with a noncontact sensor. It can be used as a contactless identification, reducing the computational resources required for low-cost and low-power consumption edge computing.

Neural Network-Based Multiplicatively Gait Feature Eradication and Detection

This research proposes an innovative method based on machine learnings for extracting and identifying gait features from multiple sources. The method aims to enhance the accuracy of gait identification by minimizing interferences caused by complex backgrounds and shelters, thereby capturing more precise information that reflects the walking characteristics of moving individuals. The technical approach involves the acquisition of gait data using a video recorder and a pyroelectric IR sensor. The image source information obtained from the video recorder is utilized to extract skeleton feature variables and Radon difference peak characteristic variables. In addition, the pyroelectric IR source information is transformed from a voltage signal to frequency domain characteristic variables. These variables are then merged after undergoing dimension reduction and signal processing. Finally, a backpropagation neural network is employed as the classifier to perform classified identification based on the merged characteristics, and the identification accuracy is evaluated. The primary application of this method is in the field of identification.

A classifier based on multiple feature extraction blocks for gait authentication using smartphone sensors

A convolutional neural network model for gait identification in the sensor domain with multiple feature extraction blocks (MFEBP) is proposed, using Residual Block and Squeeze-and-Excitation Blocks. NJUPT Datasets I, II and III have been established, containing acceleration and gyroscope signals from 113 subjects using sensors built into smartphones, in different parts of body. Dataset I is employed to pinpoint the best position of smartphone for recognition. The other two are taken to verify the performance of the designed model. The experimental results of Dataset I show that waist part is the most stable position for identification. Several experiments are also conducted to evaluate the MFEBP on OU-ISIR, ZJU, NJUPT Dataset II, III and whuGait Dataset 1, 2. The classifier with the best accuracy achieves 98.56%, 93.48%, 94.17%, 98.47%, 95.38% and 97.93% separately at a sliding length of 50 sample points. Moreover, this paper also investigates data enhancement by varying the sliding length. In the OU-ISIR dataset, the accuracy increases from 83.56% to 99.91% as the sliding length reduces from 50 to 10 sample points. In conclusion, three gait datasets are constructed and a gait model with higher recognition accuracy was proposed.

Geolocation tracking for human identification and activity recognition using radar deep transfer learning

AbstractHuman identification and activity recognition (HIAR) is crucial for many applications, such as surveillance, smart homes, and assisted living. As a sensing modality, radar has many unique characteristics including privacy protection, and contactless sensing. Single classification systems have shown to be accurate, but for long‐term solutions both human identification (ID) and human activity recognition (HAR) will need to be integrated in one system where it can be utilised simultaneously. In this article, a novel radar‐based human tracking system is presented where three classifiers are utilised to identify the subject and his/her behaviour. For any kind of motion, the system tracks the subject and detect the type of his/her motion. Based on the detected type of motion, the three classifiers are utilised for identification and activity recognition. The classifiers are built utilising deep transfer learning where three radar datasets are established to train and validate each of the deep networks. To recognise six activities and 10 human subjects, the three classifiers, namely, HAR, Gait ID, and Heart sound ID, achieve superior performance compared to the best reported results in literature with classification accuracies of 97.6%, 100%, and 41.8% respectively. Three successful examples are presented to demonstrate the introduced concept.

Design and Development of Polymer-Based Optical Fiber Sensor for GAIT Analysis

In the present scenario like COVID-19 pandemic, to maintain physical distance, the gait-based biometric is a must. Human gait identification is a very difficult process, but it is a suitable distance biometric that also gives good results at low resolution conditions even with face features that are not clear. This study describes the construction of a smart carpet that measures ground response force (GRF) and spatio-temporal gait parameters (STGP) using a polymer optical fiber sensor (POFS). The suggested carpet contains two light detection units for acquiring signals. Each unit obtains response from 10 nearby sensors. There are 20 intensity deviation sensors on a fiber. Light-emitting diodes (LED) are triggered successively, using the multiplexing approach that is being employed. Multiplexing is dependent on coupling among the LED and POFS sections. Results of walking experiments performed on the smart carpet suggested that certain parameters, including step length, stride length, cadence, and stance time, might be used to estimate the GRF and STGP. The results enable the detection of gait, including the swing phase, stance, stance length, and double supporting periods. The suggested carpet is dependable, reasonably priced equipment for gait acquisition in a variety of applications. Using the sensor data, gait recognition is performed using genetic algorithm (GA) and particle swarm optimization (PSO) technique. GA- and PSO-based gait template analyses are performed to extract the features with respect to the gait signals obtained from polymer optical gait sensors (POGS). The techniques used for classification of the obtained signals are random forest (RF) and support vector machine (SVM). The accuracy, sensitivity, and specificity results are obtained using SVM classifier and RF classifier. The results obtained using both classifiers are compared.

Ground reaction forces during double limb stances while walking in individuals with unilateral transfemoral amputation.

The asymmetrical gait of individuals with unilateral transfemoral amputation has been well documented. However, there is not a wealth of investigation into asymmetries during the double limb stance depending on whether the intact or prosthetic limb is leading. The first aim of this study was to compare ground reaction forces during the double limb stance of individuals with unilateral transfemoral amputation depending on whether their intact (initial double limb stance) or prosthetic (terminal double limb stance) limb was leading. The second aim of this study was to compare the asymmetry ratio of ground reaction forces during the double limb stance between individuals with and without unilateral transfemoral amputation. Thirty individuals, fifteen with unilateral transfemoral amputation and fifteen who were able-bodied, were recruited for this study. Each individual walked on an instrumented treadmill for 30s at eight different speeds, ranging from 2.0km/h to 5.5km/h with .5km/h increments. Ground reaction force parameters, temporal parameters, and asymmetry ratios of all parameters were computed from the data collected. The appropriate statistical analyses of all data based on normality were conducted to investigate the aims of this study. Significant main effects of speed, double limb stance, and their interactions were found for most parameters (p < .01 or p < .05). Individuals with unilateral transfemoral amputation spent a longer duration in terminal double limb stance than initial double limb stance at all tested speeds. They also experienced significantly higher peak vertical ground reaction force during initial double limb stance compared to terminal double limb stance with increasing walking speed. However, during terminal double limb stance, higher anteroposterior ground reaction force at initial contact was found when compared to initial double limb stance. Significant differences between individuals with unilateral transfemoral amputation and able-bodied individuals were found in asymmetry ratios for peak vertical ground reaction force, anteroposterior ground reaction force, anteroposterior shear, and mediolateral shear at all tested speeds. Asymmetrical loading persists in individuals with unilateral transfemoral amputation during double limb stance. Increasing walking speed increased ground reaction force loading asymmetries, which may make individuals with unilateral transfemoral amputation more susceptible to knee osteoarthritis or other musculoskeletal disorders. Further study is necessary to develop ideal gait strategies for the minimization of gait asymmetry in individuals with unilateral transfemoral amputation.

Exploration of Effective Time-Velocity Distribution for Doppler-Radar-Based Personal Gait Identification Using Deep Learning.

Personal identification based on radar gait measurement is an important application of biometric technology because it enables remote and continuous identification of people, irrespective of the lighting conditions and subjects' outfits. This study explores an effective time-velocity distribution and its relevant parameters for Doppler-radar-based personal gait identification using deep learning. Most conventional studies on radar-based gait identification used a short-time Fourier transform (STFT), which is a general method to obtain time-velocity distribution for motion recognition using Doppler radar. However, the length of the window function that controls the time and velocity resolutions of the time-velocity image was empirically selected, and several other methods for calculating high-resolution time-velocity distributions were not considered. In this study, we compared four types of representative time-velocity distributions calculated from the Doppler-radar-received signals: STFT, wavelet transform, Wigner-Ville distribution, and smoothed pseudo-Wigner-Ville distribution. In addition, the identification accuracies of various parameter settings were also investigated. We observed that the optimally tuned STFT outperformed other high-resolution distributions, and a short length of the window function in the STFT process led to a reasonable accuracy; the best identification accuracy was 99% for the identification of twenty-five test subjects. These results indicate that STFT is the optimal time-velocity distribution for gait-based personal identification using the Doppler radar, although the time and velocity resolutions of the other methods were better than those of the STFT.

Computational Intelligence Approach for Biometric Gait Identification

Gait recognition has gained significant attention in recent years due to its potential applications in various fields, including surveillance, security, and healthcare. Biometric gait identification, which involves recognizing individuals based on their walking patterns, is a challenging task due to the inherent variations in gait caused by factors such as clothing, footwear, and walking speed. In this paper, we propose a computational intelligence approach for biometric gait identification. Specifically, we integrate an intelligent convolutional model to identify human gaits based on the inertial sensory data captured from the body movement during the human walk. Extensive experiments on two datasets demonstrated that the efficiency of the proposed approach outperforms the existing methods. Our approach has the potential to be used in real-world applications such as surveillance systems and healthcare monitoring, where accurate and efficient identification of individuals based on their gait is crucial.