6 April 2022 in IEEE Sensors Journal
Using mm-wave sensing to identify persons by the way they walk has recently emerged as a promising solution of biometrics, and has found versatile applications in security surveillance, automatic access control, and health monitoring. In this paper, we build a system named MGait for indoor multi-person identification and intruder detection based on gait micro-Doppler (m-D) signatures captured by a low-cost mm-wave radar. In multi-person scenario where multiple subjects concurrently share and walk within the same physical space, the proposed system continuously detects and separately tracks each subject in the range-Doppler (R-D) space frame by frame to extract their respective gait m-D signatures. Then, an open-set identification network is trained by a large-margin Gaussian mixture (L-GM) loss to learn highly discriminative feature representations and compel the learned features of the training set to follow a Gaussian mixture distribution, with each component representing a registered user. As such, the known users can be directly identified based on the class-posterior probability and the intruder can also be rejected by setting a probability threshold. The proposed system is verified on real radar measurements collected by a 77 GHz FMCW radar, achieving an overall accuracy of 88.59% in identifying up to 5 subjects, including 1 known user and 4 intruders, freely and concurrently moving in a corridor space.
8 September 2020 in IET Radar, Sonar & Navigation
Gait‐based human identification aims to identify individuals by their walking style. In this study, the authors investigate the use of micro‐Doppler (m‐D) signatures retrieved from a frequency‐modulated continuous‐wave radar sensor to identify individuals based on their natural gait characteristics. The gait dataset of 20 persons has been collected in an indoor environment where each subject was allowed to walk naturally and freely, which is absolutely more realistic and challenging than most existing works based on limited walking behaviour. Then, they perform identification using a transfer learned ResNet‐50, which was fine‐tuned on the gait m‐D dataset based on the deep transfer learning technique. Through experiments, they first determined the optimal observation window length of m‐D samples, and with this input, they achieved an average identification accuracy of 96.7% on the test set for 20 subjects, which highly outperforms the state‐of‐the‐art methods. The presented work provides prospects in developing a solution to automatically identify persons based on gait characteristics using a simple and cost‐efficient radar device.
18 January 2021 in IEEE Sensors Journal
Human identification based on gait radar micro-Doppler signatures has received significant attention in recent years. In real world applications, a recognition system should have the ability to reject unknown identities as well as correctly recognize known identities, i.e. the open-set identification. In this paper, we propose a novel framework to tackle this challenging problem. We base our approach on a probabilistic discriminant model built on a deep discriminative representation network (DDRN). Specifically, we first train a DDRN with the cosine margin (CM) loss to map gait samples into an embedding space where the learned features belonging to the same identity are much closer together and those of different identities are further apart. Then, in the learned embedding space, the class-inclusion probability (CIP) model associated with each known identity can be constructed based on the statistical Extreme Value Theory (EVT) to bound each other's support region, which can then be used to estimate the class-belongingness probabilities of a probe sample. Finally, we threshold on these probabilities to determine whether it belongs to one of the known identities or an unknown class. Experiments on a radar measured gait dataset show the effectiveness of our approach.
9 December 2021 in IEEE Transactions on Instrumentation and Measurement
Building a robust and scalable radar-based person identification system that can accurately identify registered users and fast enroll new subjects at any time with only limited radar measurements is very desirable for real-world applications. Existing solutions, however, most resort to deep learning techniques that are heavily data-dependent and hard to scale once trained. To overcome these limitations, this article proposes a novel set-based discriminative subspaces learning (SDSL) approach and builds a high-performance gait recognition system based on it. Specifically, we model the system as a two-stage paradigm consisting of a feature embedding network for gait feature extraction, followed by an SDSL module for gait pattern modeling through low-dimensional linear subspace representing and subspaces discriminant learning. Finally, a nearest neighbor classifier (NNC) is adopted to perform person identification in the learned discriminant space by matching the subspace of a probe set with all subspaces in the gallery. Extensive experiments conducted on a radar gait dataset collected in a typical corridor scenario demonstrate not only the superiority of our model in both identification accuracy and scalability, but also its robustness against limited training data.
2 July 2019 in IEEE Access
Aiming at the problem of scattering centers resolving and angular positions estimation of spatial extended targets, a high-resolution and high-accuracy angle estimation method based on multi-task group sparse model and collocated MIMO radar is proposed, which is helpful to obtain the structure information of targets and improve the success rate of target recognition. Characterized by sparse and clustered distribution in space, angular positions estimation of multiple closely-spaced and correlated point scattering targets belonging to a spatially extended target can be modeled as a multi-task group sparse problem and can be solved by multi-task group sparse recovery. To overcome the sparse recovery performance degradation caused by high correlation in group sparse solution matrix and to improve the accuracy and robustness of angle estimation, a complex spatiotemporal sparse Bayesian learning (CST-SBL) algorithm which exploits spatiotemporal correlation structures of solution matrix is proposed to reconstruct angular positions. Compared with previous work, the proposed approach achieves high-resolution and high-accuracy estimation performance, especially in cases of low SNR and few snapshots. Theoretical analysis and simulation results validate the effectiveness of the proposed technique.