An Efficient Deep Learning Approach Using Improved Generative Adversarial Networks for Incomplete Information Completion of Self-driving Vehicles

Autonomous driving is a key technology for intelligent logistics in the Industrial Internet of Things (IIoT). In autonomous driving, the appearance of incomplete point clouds that lose geometric and semantic information is inevitable due to the limitations of occlusion, sensor resolution, and viewing angle when Light Detection And Ranging (LiDAR) is applied. The existence of incomplete point clouds, especially incomplete vehicle point clouds, would lead to a reduction in the accuracy of object detection, traffic alerts, and collision avoidance for autonomous driving vehicles. Existing point cloud completion networks, such as the Point Fractal Network (PF-Net), focus on the accuracy of point cloud completion without considering the efficiency of the inference process, which makes it difficult for them to be deployed for vehicle point cloud repair in autonomous driving. To address this problem, in this paper, we propose an efficient deep learning approach to repair incomplete vehicle point clouds in autonomous driving accurately and efficiently. In the proposed method, an efficient downsampling algorithm that combines incremental sampling and one-time sampling is presented to improve the inference speed of the PF-Net based on Generative Adversarial Network (GAN). To evaluate the performance of the proposed method, a real dataset is used, and autonomous driving scenes are created, where three incomplete vehicle point clouds with 5 different sizes are used for three autonomous driving situations. The improved PF-Net can achieve speedups of over 19x with almost the same accuracy when compared to the original PF-Net. Experimental results demonstrate that the improved PF-Net can be applied to efficiently complete vehicle point clouds in autonomous driving.