结合目标检测与双目视觉的三维车辆姿态检测

Combining Object Detection and Binocular Vision for 3D Car Pose Estimation

随着城市规模的发展,车辆的需求在与日俱增,同时对自动驾驶技术的需求也在不断提高.为了增强自动驾驶系统对路面车辆的信息掌握能力,提出一种车辆姿态检测方法.首先利用基于深度学习的目标检测方法获取车辆在二维图片上的信息,结合深度相机利用双目视觉获取车辆的关键三维空间信息;然后综合二维与三维信息建立三维空间坐标,经过计算后实现车辆的三维边框绘制,绘制的三维边框能辅助区分出车辆在空间上的方位.文中方法为端对端方法,不需要其他额外的输入信息,能够实时展示在相机中.实验结果表明,该方法针对常见的路面停车场景有较好的识别效果,对自动驾驶系统有较好的辅助作用;对比目前流行的三维边框计算方法也展示了其准确性.

With the rapid development of urban construction,the demand for vehicles is increasing day by day,which brings the demand for automatic driving technology.In order to enhance the ability of the automatic driving system to handle the information about vehicles on road,this paper proposes a car pose estimation method.The method firstly uses the object detection algorithm based on deep-learning to obtain the information of the vehicle on two-dimensional image,and then combines the depth camera with binocular vision to capture the three-dimensional spatial information of the key points of the vehicle;the 3D coordinates are built by integrating the 2D and 3D information which are used to obtain the 3D bounding box of the vehicle,the 3D bounding box could be benefit for distinguishing the car’s spatial orientation in automatic driving system.The method is an end-to-end method without any additional input,and the result can be displayed on the camera in real time.The experimental result shows that the method has a fine recognition effect for the parking lots on road and has a good auxiliary effect on the automatic driving system;the accuracy of the method is also demonstrated by comparing with the existing method.