Object Tracking and Velocity Estimation Considering Deviation Due to LiDAR Illumination Surface

For autonomous vehicles to effectively navigate urban environments where pedestrians and personal vehicles are in close proximity, it is crucial to accurately track and predict their positions and velocities to enable effective collision avoidance. Light Detection and Ranging (LiDAR) is commonly used for distance measurement due to its high accuracy and rapid sampling capabilities, and it is unaffected by ambient light. However, the point cloud generated by the laser illumination appears on the surface of the target objects, which can lead to significant velocity drift when estimating movement due to the shifting of the illuminated point cloud on these surfaces. In this study, we demonstrate that using Coherent Point Drift (CPD) with raw point cloud data allows for more accurate velocity calculations that account for fluctuations on the irradiated surfaces. Although CPD results may sometimes include spike-like measurement errors, we employ a Probabilistic Data Association Filter (PDAF) to eliminate these outliers. This approach helps track pedestrians and estimate pedestrian flow based on refined velocity estimates. Experimental validation conducted in environments with multiple pedestrians shows that these methods enhance object tracking and improve the accuracy of pedestrian flow estimation.