On Learning Disentangled Representations for Gait Recognition

Learning Gait Representation From Massive Unlabelled Walking Videos: A Benchmark

Gait depicts individuals' unique and distinguishing walking patterns and has become one of the most promising biometric features for human identification. As a fine-grained recognition task, gait recognition is easily affected by many factors and usually requires a large amount of completely annotated data that is costly and insatiable. This paper proposes a large-scale self-supervised benchmark for gait recognition with contrastive learning, aiming to learn the general gait representation from massive unlabelled walking videos for practical applications via offering informative walking priors and diverse real-world variations. Specifically, we collect a large-scale unlabelled gait dataset GaitLU-1M consisting of 1.02M walking sequences and propose a conceptually simple yet empirically powerful baseline model GaitSSB. Experimentally, we evaluate the pre-trained model on four widely-used gait benchmarks, CASIA-B, OU-MVLP, GREW and Gait3D with or without transfer learning. The unsupervised results are comparable to or even better than the early model-based and GEI-based methods. After transfer learning, GaitSSB outperforms existing methods by a large margin in most cases, and also showcases the superior generalization capacity. Further experiments indicate that the pre-training can save about 50% and 80% annotation costs of GREW and Gait3D. Theoretically, we discuss the critical issues for gait-specific contrastive framework and present some insights for further study. As far as we know, GaitLU-1M is the first large-scale unlabelled gait dataset, and GaitSSB is the first method that achieves remarkable unsupervised results on the aforementioned benchmarks.

GaitSG: Gait Recognition with SMPLs in Graph Structure

Gait recognition aims to identify a person based on his unique walking pattern. Compared with silhouettes and skeletons, skinned multi-person linear (SMPL) models can simultaneously provide human pose and shape information and are robust to viewpoint and clothing variances. However, previous approaches have only considered SMPL parameters as a whole and are yet to explore their potential for gait recognition thoroughly. To address this problem, we concentrate on SMPL representations and propose a novel SMPL-based method named GaitSG for gait recognition, which takes SMPL parameters in the graph structure as input. Specifically, we represent the SMPL model as graph nodes and employ graph convolution techniques to effectively model the human model topology and generate discriminative gait features. Further, we utilize prior knowledge of the human body and elaborately design a novel part graph pooling block, PGPB, to encode viewpoint information explicitly. The PGPB also alleviates the physical distance-unaware limitation of the graph structure. Comprehensive experiments on public gait recognition datasets, Gait3D and CASIA-B, demonstrate that GaitSG can achieve better performance and faster convergence than existing model-based approaches. Specifically, compared with the baseline SMPLGait (3D only), our model achieves approximately twice the Rank-1 accuracy and requires three times fewer training iterations on Gait3D.

Condition-Adaptive Graph Convolution Learning for Skeleton-Based Gait Recognition

Graph convolutional networks have been widely applied in skeleton-based gait recognition. A key challenge in this task is to distinguish the individual walking styles of different subjects across various views. Existing state-of-the-art methods employ uniform convolutions to extract features from diverse sequences and ignore the effects of viewpoint changes. To overcome these limitations, we propose a condition-adaptive graph (CAG) convolution network that can dynamically adapt to the specific attributes of each skeleton sequence and the corresponding view angle. In contrast to using fixed weights for all joints and sequences, we introduce a joint-specific filter learning (JSFL) module in the CAG method, which produces sequence-adaptive filters at the joint level. The adaptive filters capture fine-grained patterns that are unique to each joint, enabling the extraction of diverse spatial-temporal information about body parts. Additionally, we design a view-adaptive topology learning (VATL) module that generates adaptive graph topologies. These graph topologies are used to correlate the joints adaptively according to the specific view conditions. Thus, CAG can simultaneously adjust to various walking styles and viewpoints. Experiments on the two most widely used datasets (i.e., CASIA-B and OU-MVLP) show that CAG surpasses all previous skeleton-based methods. Moreover, the recognition performance can be enhanced by simply combining CAG with appearance-based methods, demonstrating the ability of CAG to provide useful complementary information.

Person Recognition Based on Deep Gait: A Survey

Gait recognition, also known as walking pattern recognition, has expressed deep interest in the computer vision and biometrics community due to its potential to identify individuals from a distance. It has attracted increasing attention due to its potential applications and non-invasive nature. Since 2014, deep learning approaches have shown promising results in gait recognition by automatically extracting features. However, recognizing gait accurately is challenging due to the covariate factors, complexity and variability of environments, and human body representations. This paper provides a comprehensive overview of the advancements made in this field along with the challenges and limitations associated with deep learning methods. For that, it initially examines the various gait datasets used in the literature review and analyzes the performance of state-of-the-art techniques. After that, a taxonomy of deep learning methods is presented to characterize and organize the research landscape in this field. Furthermore, the taxonomy highlights the basic limitations of deep learning methods in the context of gait recognition. The paper is concluded by focusing on the present challenges and suggesting several research directions to improve the performance of gait recognition in the future.

Model-based and model-free deep features fusion for high performed human gait recognition

In the last decade, the need for a non-contact biometric model for recognizing candidates has increased, especially after the pandemic of COVID-19 appeared and spread worldwide. This paper presents a novel deep convolutional neural network (CNN) model that guarantees quick, safe, and precise human authentication via their poses and walking style. The concatenated fusion between the proposed CNN and a fully connected model has been formulated, utilized, and tested. The proposed CNN extracts the human features from two main sources: (1) human silhouette images according to model-free and (2) human joints, limbs, and static joint distances according to a model-based via a novel, fully connected deep-layer structure. The most commonly used dataset, CASIA gait families, has been utilized and tested. Numerous performance metrics have been evaluated to measure the system quality, including accuracy, specificity, sensitivity, false negative rate, and training time. Experimental results reveal that the proposed model can enhance recognition performance in a superior manner compared with the latest state-of-the-art studies. Moreover, the suggested system introduces a robust real-time authentication with any covariate conditions, scoring 99.8% and 99.6% accuracy in identifying casia (B) and casia (A) datasets, respectively.

Deep Gait Recognition: A Survey

Gait recognition is an appealing biometric modality which aims to identify individuals based on the way they walk. Deep learning has reshaped the research landscape in this area since 2015 through the ability to automatically learn discriminative representations. Gait recognition methods based on deep learning now dominate the state-of-the-art in the field and have fostered real-world applications. In this paper, we present a comprehensive overview of breakthroughs and recent developments in gait recognition with deep learning, and cover broad topics including datasets, test protocols, state-of-the-art solutions, challenges, and future research directions. We first review the commonly used gait datasets along with the principles designed for evaluating them. We then propose a novel taxonomy made up of four separate dimensions namely body representation, temporal representation, feature representation, and neural architecture, to help characterize and organize the research landscape and literature in this area. Following our proposed taxonomy, a comprehensive survey of gait recognition methods using deep learning is presented with discussions on their performances, characteristics, advantages, and limitations. We conclude this survey with a discussion on current challenges and mention a number of promising directions for future research in gait recognition.

Learning Gait Representations with Noisy Multi-Task Learning

Gait analysis is proven to be a reliable way to perform person identification without relying on subject cooperation. Walking is a biometric that does not significantly change in short periods of time and can be regarded as unique to each person. So far, the study of gait analysis focused mostly on identification and demographics estimation, without considering many of the pedestrian attributes that appearance-based methods rely on. In this work, alongside gait-based person identification, we explore pedestrian attribute identification solely from movement patterns. We propose DenseGait, the largest dataset for pretraining gait analysis systems containing 217 K anonymized tracklets, annotated automatically with 42 appearance attributes. DenseGait is constructed by automatically processing video streams and offers the full array of gait covariates present in the real world. We make the dataset available to the research community. Additionally, we propose GaitFormer, a transformer-based model that after pretraining in a multi-task fashion on DenseGait, achieves 92.5% accuracy on CASIA-B and 85.33% on FVG, without utilizing any manually annotated data. This corresponds to a +14.2% and +9.67% accuracy increase compared to similar methods. Moreover, GaitFormer is able to accurately identify gender information and a multitude of appearance attributes utilizing only movement patterns. The code to reproduce the experiments is made publicly.

Advances in Vision-Based Gait Recognition: From Handcrafted to Deep Learning

Identifying people’s identity by using behavioral biometrics has attracted many researchers’ attention in the biometrics industry. Gait is a behavioral trait, whereby an individual is identified based on their walking style. Over the years, gait recognition has been performed by using handcrafted approaches. However, due to several covariates’ effects, the competence of the approach has been compromised. Deep learning is an emerging algorithm in the biometrics field, which has the capability to tackle the covariates and produce highly accurate results. In this paper, a comprehensive overview of the existing deep learning-based gait recognition approach is presented. In addition, a summary of the performance of the approach on different gait datasets is provided.

Robust clothing-independent gait recognition using hybrid part-based gait features

Recently, gait has been gathering extensive interest for the non-fungible position in applications. Although various methods have been proposed for gait recognition, most of them can only attain an excellent recognition performance when the probe and gallery gaits are in a similar condition. Once external factors (e.g., clothing variations) influence people's gaits and changes happen in human appearances, a significant performance degradation occurs. Hence, in our article, a robust hybrid part-based spatio-temporal feature learning method is proposed for gait recognition to handle this cloth-changing problem. First, human bodies are segmented into the affected and non/less unaffected parts based on the anatomical studies. Then, a well-designed network is proposed in our method to formulate our required hybrid features from the non/less unaffected body parts. This network contains three sub-networks, aiming to generate features independently. Each sub-network emphasizes individual aspects of gait, hence an effective hybrid gait feature can be created through their concatenation. In addition, temporal information can be used as complement to enhance the recognition performance, a sub-network is specifically proposed to establish the temporal relationship between consecutive short-range frames. Also, since local features are more discriminative than global features in gait recognition, in this network a sub-network is specifically proposed to generate features of local refined differences. The effectiveness of our proposed method has been evaluated by experiments on the CASIA Gait Dataset B and OU-ISIR Treadmill Gait Dataset B. Related experiments illustrate that compared with other gait recognition methods, our proposed method can achieve a prominent result when handling this cloth-changing gait recognition problem.

A Unified Local–Global Feature Extraction Network for Human Gait Recognition Using Smartphone Sensors

Smartphone-based gait recognition has been considered a unique and promising technique for biometric-based identification. It is integrated with multiple sensors to collect inertial data while a person walks. However, captured data may be affected by several covariate factors due to variations of gait sequences such as holding loads, wearing types, shoe types, etc. Recent gait recognition approaches either work on global or local features, causing failure to handle these covariate-based features. To address these issues, a novel weighted multi-scale CNN (WMsCNN) architecture is designed to extract local to global features for boosting recognition accuracy. Specifically, a weight update sub-network (Ws) is proposed to increase or reduce the weights of features concerning their contribution to the final classification task. Thus, the sensitivity of these features toward the covariate factors decreases using the weight updated technique. Later, these features are fed to a fusion module used to produce global features for the overall classification. Extensive experiments have been conducted on four different benchmark datasets, and the demonstrated results of the proposed model are superior to other state-of-the-art deep learning approaches.

Reliability and generalization of gait biometrics using 3D inertial sensor data and 3D optical system trajectories

Particularities in the individuals’ style of walking have been explored for at least three decades as a biometric trait, empowering the automatic gait recognition field. Whereas gait recognition works usually focus on improving end-to-end performance measures, this work aims at understanding which individuals’ traces are more relevant to improve subjects’ separability. For such, a manifold projection technique and a multi-sensor gait dataset were adopted to investigate the impact of each data source characteristics on this separability. Assessments have shown it is hard to distinguish individuals based only on their walking patterns in a subject-based identification scenario. In this setup, the subjects’ separability is more related to their physical characteristics than their movements related to gait cycles and biomechanical events. However, this study’s results also points to the feasibility of learning identity characteristics from individuals’ walking patterns learned from similarities and differences between subjects in a verification setup. The explorations concluded that periodic components occurring in frequencies between 6 and 10 Hz are more significant for learning these patterns than events and other biomechanical movements related to the gait cycle, as usually explored in the literature.