People Detection and Tracking Using Ground LiDAR

Abstract

People detection and tracking are crucial issues in various fields, such as surveillance, security, and intelligent transportation systems. This paper presents a people detection and tracking method using light detection and ranging (LiDAR) set in an environment. People detection is achieved using a one-dimensional convolutional neural network (1D-CNN) together with the background subtraction method. Regions of interest are detected based on the background subtraction method, and people are detected in those regions using 1D-CNN. The detected people are tracked using the interacting multimodel estimator; people positions, velocities, and behaviors, such as stopping, walking, and suddenly rushing out, are estimated. Simulation and real-world experiments are conducted using a Velodyne 32-layer LiDAR. The experimental results show that the people tracker conjunction with people detection using both the 1D-CNN and background subtraction method enables accurate multipeople tracking.