LiDAR Point Cloud Data Processing in Autonomous Vehicles

Abstract

In Autonomous vehicles, LiDAR point cloud data is an important source to identify different obstacles present in the environment. 3D LiDAR point cloud data has high density, outlier noise and scattered distribution in a freeway road scene, which is not conducive for the ground point segmentation process. The point cloud data preprocessing is essential for the frame matching of the different sequences of the road scene. It also improves the computational efficiency and storage capacity of the system. This paper shows a three-step novel methodology for implementing in real-time. In the first step, filtered the data points using linear interpolation, the second step, the outliers are removed using the statistical method, and in the final step, downsampled the LiDAR data points using voxel grid filters. The experiment result shows that the data volume is reduced by 50% without losing any spatial information.