Gait Recognition Using Spatio-Temporal Information of 3D Point Cloud via Millimeter Wave Radar

Gait recognition is one of the crucial methods in identity recognition, which has a wide range of applications in many fields, such as smart home, smart office, and health monitoring. The camera is the most mainstream traditional solution. But the camera is difficult to maintain stable performance in the dark, low light, and bad weather conditions. In addition, privacy leakage is also one of the important issues that people worry about. In contrast, as the latest research progress in gait recognition, millimeter wave radar can not only protect people's privacy, but also maintain normal perception performance in dark conditions. In this paper, we propose a system for gait recognition named MTPGait using spatio-temporal information via millimeter wave radar. We specially design a neural network that can extract multiscale spatio-temporal features along space and time dimensions of 3D point cloud concisely and efficiently. We use LSTM to design the context flow of local and global time and space, fusing local and global spatio-temporal features. In addition, we construct and release a millimeter wave radar 3D point cloud data set, which consists of 960-minute gait data of 40 volunteers. Using the data set, we evaluate the system and compare it with four state-of-the-art algorithms. The experimental results show that MTPGait is able to achieve 96.7% recognition accuracy in a single-person scene on fixed route and 90.2% recognition accuracy when two people coexist, while none of the existing methods is more than 90% recognition accuracy in either scenario.