Denoising Framework Based on Multiframe Continuous Point Clouds for Autonomous Driving LiDAR in Snowy Weather

Abstract

Adverse weather conditions are one of the long-tailed problems facing autonomous driving. Solving the problem of autonomous driving operation in adverse weather conditions is an important challenge for realizing advanced autonomous driving. To enhance the LiDAR perception capability in snowy weather for autonomous driving, this study proposes a denoising method for multiframe continuous point clouds. The core concept of this method is to allow ordered objects (e.g., stationary objects on the ground) to strengthen each other while allowing disordered objects (e.g., snow) to weaken each other. This is done by first selecting three consecutive frames of the point cloud as a denoising unit and then removing the ground points from each frame of the point cloud. After that, the point clouds from the first two frames are used as the source point clouds, and the point cloud from the third frame is used as the target point cloud for point cloud registration. Finally, the time outlier removal (TOR) filter proposed in this article combined with the entropy weight method (EWM) is utilized for denoising. The experimental results show that the performance of the method proposed in this article exceeds the existing methods. In addition, the method in this article not only removes the disordered snowflakes in the air, but also removes some other disordered noise points (e.g., the ghosting of the stationary objects), which provides an advantageous guarantee for the realization of automatic driving in snowy weather.

Topics & Concepts