Abstract
Gait recognition, which aims at identifying individuals by their walking patterns, has achieved great success based on silhouette. The binary silhouette sequence encodes the walking pattern within the sparse boundary representation. Therefore, most pixels in the silhouette are under-sensitive to the walking pattern since the sparse boundary lacks dense spatial-temporal information, which is suitable to be represented with dense texture. To enhance the sensitivity to the walking pattern while maintaining the robustness of recognition, we present a Complementary Learning with neural Architecture Search (CLASH) framework, consisting of walking pattern sensitive gait descriptor named dense spatial-temporal field (DSTF) and neural architecture search based complementary learning (NCL). Specifically, DSTF transforms the representation from the sparse binary boundary into the dense distance-based texture, which is sensitive to the walking pattern at the pixel level. Further, NCL presents a task-specific search space for complementary learning, which mutually complements the sensitivity of DSTF and the robustness of the silhouette to represent the walking pattern effectively. Extensive experiments demonstrate the effectiveness of the proposed methods under both in-the-lab and in-the-wild scenarios. On CASIA-B, we achieve rank-1 accuracy of 98.8%, 96.5%, and 89.3% under three conditions. On OU-MVLP, we achieve rank-1 accuracy of 91.9%. Under the latest in-the-wild datasets, we outperform the latest silhouette-based methods by 16.3% and 19.7% on Gait3D and GREW, respectively.
Abstract (translated)
基于轮廓识别,通过识别行走模式来识别个体,已经在很大程度上取得了成功。二进制轮廓序列根据轮廓稀疏表示编码了行走模式。因此,轮廓中的大多数像素对于行走模式都比较不敏感,因为稀疏边界缺乏密集的空间-时间信息,而密集纹理可以表示这种信息。为了在保持识别的鲁棒性的同时提高对行走模式的敏感性,我们提出了一个互补学习与神经网络架构搜索(CLASH)框架。具体来说,DSTF将二进制轮廓表示从稀疏二进制边界转换为密集距离纹理,对像素层面上的行走模式非常敏感。此外,NCL为互补学习任务提供了一个任务特定的搜索空间,有效地互补DSTF的敏感性和轮廓对行走模式的鲁棒性,使其更加有效地表示行走模式。在广泛的实验中,我们证明了所提出方法在实验室和野外场景中的有效性。在CASIA-B上,我们实现了98.8%,96.5%和89.3%的排名1精度。在OU-MVLP上,我们实现了91.9%的排名1精度。在最新的野外数据集上,我们在Gait3D和GREW上分别比基于轮廓的方法提高了16.3%和19.7%的性能。
URL
https://arxiv.org/abs/2407.03632