采用桌面视觉系统实现深度测量的方法

Depth Measurement Based on Desktop Vision System

提出了一种基于桌面视觉系统(DVS)的深度测量方法,并推导了相关的立体图像对校正方法和计算公式,主要包括:几何畸变校正、中心投影的重投影变换、缩放变换、竖直平移变换以及亮度和颜色校正等。通过重投影变换可以使内参数彼此不一致的两个会聚式相机拍摄得到的立体图像对,变换为平行式摆放且具有相同内参数(除几何畸变系数外)的两相机拍摄得到的立体图像对。该方法首先对桌面视觉系统拍摄的立体图像对进行校正处理,使其变换为理想的立体图像对;然后,对校正后的立体图像对进行立体匹配,得到各对同名点的水平视差;最后,基于水平视差和摄像机标定的参数,应用三角测量法求空间点的深度信息。实验结果表明,测量方法的相对误差在0.31%以内。

An effective depth measurement method based on desktop vision system （DVS） with high precision is proposed. For this purpose, rectification techniques for stereo image pairs are proposed and derived, which mainly consist of geometric distortion correction, reprojection transformation by central projection, zooming transformation, vertical translation transformation and brightness and color correction. The reprojection transformation can change the stereo image pairs captured by a binocular converged camera system with different camera intrinsic parameters into those captured by binocular parallel camera system with the same camera intrinsic parameters （except geometric distortion coefficient）. For depth measurement, firstly, the stereo image pairs captured by a DVS are rectified and corrected into perfect ones. And then, horizontal parallax of the corresponding points of interest is gotten by stereo matching. Finally, the depth information of the spatial point of interest is calculated based on horizontal parallax and the parameters obtained by camera calibration. Experimental results show that the proposed depth measurement method is simple and practical, and the relative error is demonstrated for depth measurement of the near distant objects to be no more than 0.31%.