Vehicle-Mounted LiDAR Multiobject Detection and Tracking Under Weak Point Cloud

LiDAR is a key sensor for high-level self-driving cars to sense the road environment. However, LiDAR sensors suffer from weak point clouds due to the absorption and diffraction of rain and snow particles in rainy and snowy weather conditions, which reduces the accuracy and reliability of object detection and tracking. To address the shortcomings of LiDAR sensors in weak point cloud scenarios, this article proposes a multiobject detection and tracking strategy based on regional feature enhancement, in which more detailed local features are obtained by reconstructing the perception of weak point cloud scenes and optimizing the scene coding from voxels to key points. To address the issue of unstable tracking under weak point cloud features, a multiobject tracking (MOT) method is proposed. This method integrates a multicategory tracking module with an unscented Kalman filter (UKF) to optimize motion prediction and updating processes. Additionally, an adaptive weak point cloud factor is introduced during the noise update, resulting in more accurate and stable MOT under weak point cloud features working conditions. Simulation results demonstrate that the AMOTA of the proposed method reaches 74.40%, which exceeds the mainstream methods, Poly-MOT and Fast-Poly.