基于三维模型的单目车辆位姿估计  

Monocular vehicle pose estimation based on 3D model

作　　者：许凌志 符钦伟 陶卫[1] 赵辉[1] XU Ling-zhi;FU Qin-wei;TAO Wei;ZHAO Hui(Department of Instrument Science and Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学仪器科学与工程系,上海200240

出　　处：《光学精密工程》2021年第6期1346-1355,共10页Optics and Precision Engineering

基　　金：国家重点研发计划资助项目(No.2018YFB1305005)。

摘　　要：车辆位姿估计是智慧交通系统的重要组成部分,然而复杂的运动场景以及单目相机存在无法获取深度信息等问题。本文提出了一种结合单目相机及车辆三维模型进行车辆位姿估计的方法。首先对多尺度的车辆目标进行尺度归一化,然后以向量场的形式回归车辆关键点的坐标提升在遮挡或者截断状态下的位姿估计精度。在此过程中提出使用基于距离加权的向量场损失函数和关键点误差最小化的投票方法,进一步提高了位姿估计算法的准确性。此外,本文制作了一个含有丰富标注信息的合成车辆位姿估计数据集,在其上的验证结果表明,本文算法的平均定位误差和角度误差分别为0.162 m和4.692°,在实际场景中有着非常大的应用价值。Vehicle pose estimation is an important component of intelligent transportation systems.How⁃ever,the complex scenes and loss of depth information are challenging problems in the estimation.This paper proposes a method that combines monocular pose estimation and a 3D vehicle model to estimate ve⁃hicle pose.First,a multi-scale vehicle are normalized,and then the coordinates of key points are predicted in the form of a vector field to increase the accuracy of the pose estimation for the truncated and occluded vehicle.Furthermore,a distance-based loss function for the vector field and key point error minimization voting method is established to further improve the accuracy of the pose estimation algorithm.In addition,we propose a synthetic vehicle pose estimation dataset with rich annotation information.The verification results show that the average position and angle errors of our algorithm are 0.162 m and 4.692°,respec⁃tively.Our method provides significant improvements over existing methods and has considerable practical application value.

关 键 词：单目视觉 车辆位姿 三维模型 向量场 关键点 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP277[自动化与计算机技术—控制科学与工程]