A Quality Metric for 3D LiDAR Point Cloud based on Vision Tasks

In this paper, we present an adaptive point cloud compression implementation scheme based on machine vision tasks. In order to make this scheme more effective, we also provide an objective quality assessment metric of 3D LiDAR point cloud. In the field of point cloud quality assessment, it is currently the first assessment model which focus on machine vision tasks. Considering the characteristics of LiDAR point cloud data, the metric is computed based on direction representation of local point set and density information of spherical region. Finally, we evaluate its performance on open LiDAR point cloud dataset, KITTI, and the two compression platforms, TMC13 and CTIV, are used to get the distorted data. Based on the distorted data, we calculated both the metric and the accuracy of vision task. The experimental result shows that we get a strong linear correlation between this metric and the vision task accuracy, which outperforms its counterparts.