3D LiDAR Automatic Driving Environment Detection System Based on MobileNetv3-YOLOv4

Abstract

In this paper, we proposed 3D LiDAR Automatic Driving Environment Detection System Based on MobileNetv3-YOLOv4. In recent years, artificial intelligence and automatic driving technology have developed very rapidly. Automatic driving has the advantages of law-abiding and fast response, which can significantly reduce driver and passenger casualties. However, due to the large number of parameters and complexity of most object detection neural networks, the computation time required is huge. To solve this problem, this paper applies the lightweight technique of Mobilenetv3 to significantly improve the original object detection neural network, and finds the region of interest by using point cloud de-grounding and clustering algorithms. The data from the region of interest is fed into the Mobilenetv3-YOLOv4 neural network for detection to perform the high accuracy of object detection.