Vehicle Detection and Shape Refinement Based on LiDAR

Abstract

Nowadays, the vehicle detection technology based on lidar has been widely used in the field of intelligent connected vehicles. Accurate vehicle detection and shape estimation can provide accurate static information for follow-up tracking and prediction. Although the existing vehicle detection algorithm based on lidar can accurately detect and classify each target vehicle, there is a problem of unstable estimation when the vehicle point cloud contour is partially obscured in an "L" shape. In order to solve this problem, this paper proposes a vehicle shape optimization algorithm based on point cloud cluster features, the proposed algorithm outputs vehicle detection results with PointPillars target detection algorithm. Vehicle shape refinement module based on point cloud cluster features is adopted to further estimate the shape and heading of vehicle detection results. The experimental results in multiple scenes show that the vehicle shape refinement algorithm proposed in this paper can effectively improve the stability of vehicle shape and heading estimation to a certain extent compared with the existing vehicle detection algorithms, and the average consuming-time per frame of all algorithm module is 88.93 ms, which can meet the real-time requirement of vehicles.

Keywords

• intelligent connected vehicle
• environmental perception
• lidar
• vehicle detection
• shape refinement