自主车辆视觉系统的摄像机动态自标定算法

Dynamic Approach of Camera Auto-Calibration for Vision System on Autonomous Vehicle

为了解决自主车辆视觉传感器在姿态变化后摄像机旋转外参数的在线自标定问题,提出了一种利用道路图像消失点相对不变量的摄像机旋转外参数的动态自标定算法.它首先从汽车采集到的道路视频序列中实时分析道路消失点,然后依据中心极限定理并使用相应的统计量,动态地估计主消失点与消失线参数,最后利用小孔成像模型下的消失线与消失点方程,解析地求得对应于当前统计特征的摄像机外部参数动态解.实验结果表明,在摄像机进行姿态调整后,实验系统使用本算法可在不超过90帧有效图像样本的基础上,能对摄像机旋转外参数进行精确标定,使旋转外参数的绝对误差值收敛于0.001 rad的误差带之内.

A dynamic auto-calibration approach using relevant invariant of vanishing points in the road image was presented in order to solve the on-line auto-calibration problem of extrinsic camera rotation parameters after the autonomous vehicle changes the pose of the vision sensors. Firstly the vanishing point was analyzed in real time from the road video sequence collected by the car, then the main vanishing point and vanishing line were dynamically estimated according to the central limit theorem and using the corresponding statistic, and finally the dynamic solution that corresponds to camera extrinsic parameters of the current statistical characteristics was obtained by analytically solving the vanishing point and line equations with the pinhole imaging model. The experiment shows that after adjusting the pose of the camera, this algorithm can precisely calibrate the extrinsic rotation parameters within ninety frames and the absolute error of extrinsic rotation parameters is less than 0. 001 rad.