A Variable Dimension-Based Method for Roadside LiDAR Background Filtering

How to exclude the irrelevant information (background points) from the roadside LiDAR data and keep the objects of interest (e.g., vehicles and pedestrians) under congestions and occlusions is a challenge for LiDAR data processing. This paper developed a computation-effective point-based method for the roadside LiDAR data background filtering. The proposed method can provide a dynamic matrix to store the locations of the background points. The background points were identified based on the number of the neighbors, and the distance between the points in the current frame and the aggregated frames. Performance of the proposed method was evaluated using the LiDAR data collected at different scenarios. The results showed that the proposed method could effectively filter the background points under different situations (roads with different speed limits, congested traffic, and package loss, etc.). Compared to the state-of-the-art of filtering, the proposed method could reduce the computation load and improve the accuracy of background filtering.