Background Extraction and Objects Segmentation with 3D Roadside LiDAR under Snowy Weather

This paper proposes a new method to extract background and segment targets from point clouds collected by three-dimensional roadside LiDAR in snowfall weather. Background point extraction and target segmentation are two main problems in environmental perception based on roadside LiDAR. This paper first introduces a new background filtering algorithm, which uses the historical point cloud sequence to construct the background model in real time and filter the background by background difference. Then, a non-background target segmentation algorithm is proposed, including a linear density filter for filtering snow noise and a hierarchical target clustering algorithm at the beam level, which effectively realizes the accurate detection of road targets affected by snow occlusion and avoids over-segmentation and under-segmentation. Finally, the performance of our method is compared with that of the commonly used background extraction and target segmentation methods by using the data collected in the snowfall environment by our intelligent roadside system. The results show that the target detection accuracy rate and recall rate of our method can reach 95.5% and 96.7%, respectively, which are better than those of the existing methods, and can be more accurate and effective to segment road targets from the snowfall cloud.

I. Introduction

Automatic driving is the critical technology to support the development of a new generation of intelligent transportation in the future. It is also the inevitable trend of automobile development in the future [1]. The collaborative model of “smart car + smart road” is the most effective means to ensure the safe operation of autonomous vehicles. With the help of the regional collaborative perception advantage of intelligent roadside infrastructure, it can make up for the deficiency of single-vehicle intelligent local perception ability [2].