融合深度卷积网络与点云网络的三维车辆检测方法分析

在常规的车辆目标检测中,YOLO,SSD,RCNN等深度模型都获得了较好的检测效果,但是在无人驾驶系统中,车辆的速度、方向、相对距离等因素对于系统来说十分重要,所以采用二维车辆检测对于驾驶场景的理解还远远不够。激光点云数据蕴含着丰富的三维环境信息,融合点云数据和深度网络的三维车辆检测已成为未来的发展方向。文章给出了一种基于点云网络与卷积神经网络的三维车辆检测方法,首先,使用CRC和输入尺寸有关的SDP技术来提高车辆检测的准确性;其次,采用点云网络结构(Pointnet)来处理点云数据,实现三维目标检测,研究表明设计网络结构在检测精度上有着较大的优势。

Analysis on 3D vehicle detection with point cloud network and convolutional neural network

In the two-dimensional vehicle target detection,YOLO,SSD,RCNN and other deep network models have obtained good detection results,but in vehicle detection,the speed,direction,relative distance and other factors of the vehicle are very important for the unmanned system.Important,it is not enough to use a two-dimensional vehicle to detect real 3D driving scenes.Point cloud data contains rich 3D environment information.Using point cloud data and deep network to describe 3D scenes is the key to solving 3D vehicle detection.In this paper,a 3D vehicle detection method combining point cloud network and convolutional neural network is designed.This method first use scale set and layer-by-layer cascade rejection.CRC and SDP related to input size to improve the detection results of target vehicles,then uses point cloud network structure(Pointnet)to process point cloud data and detects the 3D vehicle.Experiments prove that our network structure has great advantages in accuracy.