GaitSet: Regarding Gait as a Set for Cross-View Gait Recognition

As a unique biometric feature that can be recognized at a distance, gait hasbroad applications in crime prevention, forensic identification and socialsecurity. To portray a gait, existing gait recognition methods utilize either agait template, where temporal information is hard to preserve, or a gaitsequence, which must keep unnecessary sequential constraints and thus loses theflexibility of gait recognition. In this paper we present a novel perspective,where a gait is regarded as a set consisting of independent frames. We proposea new network named GaitSet to learn identity information from the set. Basedon the set perspective, our method is immune to permutation of frames, and cannaturally integrate frames from different videos which have been filmed underdifferent scenarios, such as diverse viewing angles, different clothes/carryingconditions. Experiments show that under normal walking conditions, oursingle-model method achieves an average rank-1 accuracy of 95.0% on the CASIA-Bgait dataset and an 87.1% accuracy on the OU-MVLP gait dataset. These resultsrepresent new state-of-the-art recognition accuracy. On various complexscenarios, our model exhibits a significant level of robustness. It achievesaccuracies of 87.2% and 70.4% on CASIA-B under bag-carrying and coat-wearingwalking conditions, respectively. These outperform the existing best methods bya large margin. The method presented can also achieve a satisfactory accuracywith a small number of frames in a test sample, e.g., 82.5% on CASIA-B withonly 7 frames. The source code has been released athttps://github.com/AbnerHqC/GaitSet.