Vehicle Object Detection Based on RGB-Camera and Radar Sensor Fusion

Abstract

Object detection has always played a major role in the field of computer vision. With the development of faster region-based convolutional neural network (faster-RCNN), single shot multi-box detector (SSD), You Only Look Once (Yolo) and other deep learning networks, 2D object detection has been developed well developed. However, in some key areas, such as driverless vehicles and artificial intelligence robots, cannot meet the requirements of tasks just by detecting two dimensional (2D) objects. So we need to get the three dimensional (3D) position of the object. In 3D object detection, it is difficult to obtain object position information only by red-green-blue (RGB) image. There are some methods to get the 3D position of the target through the point cloud data set. Therefore, this paper proposes a deep learning network, which can combine both color image information and point cloud geometry information, while using MobileNet V3 as the backbone network greatly enhances the operation speed. The region proposal networks (RPN) layer in faster-RCNN is migrated to the network. After testing, the method proposed by us improves obviously in speed and recognition accuracy compared with the previous methods.