SpheriGait: Enriching Spatial Representation via Spherical Projection for LiDAR-based Gait Recognition

TL;DR

A method named SpheriGait is proposed for extracting and enhancing dynamic features from point clouds for Lidar-based gait recognition, which substitutes the conventional point cloud plane projection method with spherical projection to augment the perception of dynamic feature.

Abstract

Gait recognition is a rapidly progressing technique for the remote identification of individuals. Prior research predominantly employing 2D sensors to gather gait data has achieved notable advancements; nonetheless, they have unavoidably neglected the influence of 3D dynamic characteristics on recognition. Gait recognition utilizing LiDAR 3D point clouds not only directly captures 3D spatial features but also diminishes the impact of lighting conditions while ensuring privacy protection.The essence of the problem lies in how to effectively extract discriminative 3D dynamic representation from point clouds.In this paper, we proposes a method named SpheriGait for extracting and enhancing dynamic features from point clouds for Lidar-based gait recognition. Specifically, it substitutes the conventional point cloud plane projection method with spherical projection to augment the perception of dynamic feature.Additionally, a network block named DAM-L is proposed to extract gait cues from the projected point cloud data. We conducted extensive experiments and the results demonstrated the SpheriGait achieved state-of-the-art performance on the SUSTech1K dataset, and verified that the spherical projection method can serve as a universal data preprocessing technique to enhance the performance of other LiDAR-based gait recognition methods, exhibiting exceptional flexibility and practicality.

Figures (5)

• Figure 1: Plane projection and spherical projection.
• Figure 2: The influence of the z-axis coordinate and radius of the reference sphere on the projected depth map
• Figure 3: (a)We obtain gait depth maps that enhance dynamic features by projecting 3D point cloud spheres. Then, the feature map is obtained using four stages including three layers of DAM-L blocks. Finally, use Temporary Pooling and Horizontal Pooling Matching to extract features and calculate Triplet Loss and Cross Entropy Loss.(b)The DAM-L block utilizes two convolutional network branches to extract static and dynamic features, respectively.
• Figure 4: The influence of reference spheres with varying (a) z-axis coordinates and (b) radii on the identification performance of SpheriGait. The figure displays the recognition accuracy of plane projection and the top three reference sphere projections in each category, with both achieving optimal accuracy at positions slightly above the center height and slightly below the average radius, respectively.
• Figure 5: (a) Comparing the accuracy of spherical projection and planar projection under the same convolutional network recognition method, it can be found that the spherical projection of the reference sphere $z=c$ has the best overall accuracy.(b) Under the same Teansformer recognition method, it can be found that the spherical projection of the reference sphere $z=c+l$ has the best overall accuracy.