A Perception-Prediction System Design Using 3D Point Clouds

Robot safety depends in part on the ability to detect moving objects and predict their motions. Although various related algorithms have been proposed, both the real-time performance and robustness of perception and prediction still need to be improved. This paper designs a real-time and reliable perception-prediction system to perceive the obstacle and predict trajectories by using 3D point clouds. The perception-prediction system includes five modules, which preserve the low-computational characteristic. A compensation module is proposed to improve the tracking performance when an obstacle is perceived. Moreover, a stabilization module including spline interpolation, filtering, and Gaussian noise is provided to improve the prediction accuracy. Finally, experimental results are given to demonstrate the effectiveness and advantages of the proposed perception-prediction system.