动态图像逆透视映射方法在路面图像拼接中的应用

Application of Dynamic Inverse Perspective Mapping in Road Surface Image Mosaic

目前广泛使用的车载路面图像逆透视映射方法均假定相机姿态角是恒定的。车辆行进中的颠簸会使得相机姿态发生变化,由此导致现有方法获得的路面俯视图出现较大误差。为此,本文使用姿态传感器实时测量相机姿态,利用相机姿态角数据计算出相机相对路面的逆透视映射关系矩阵,从而获得了高精度的路面俯视图。实验测试结果表明,在相机任意姿态角下,由该方法获得的俯视图其逆透视映射误差不超过0.8%。室外动态颠簸状况下的路面图像拼接实验表明,和传统方法相比,该方法获得的路面俯视图失真较小。因此,改进后的方法可在相机处于颠簸条件下获得高精度的路面俯视图,可应用于机场跑道异物快速检测和公路路面病害勘测等领域。

The inverse perspective mapping （IPM）algorithm existing can only be used to the condition that the camera attitude does not change. The balleting of the vehicle during moving will bring variation to the camera attitude, and the IPM algorithm existing will induce error. Consequently, a real time IPM method is proposed. This paper used attitude sensor to measure the camera attitude in real time, and deduces the relational matrix between camera coordinates and the road surface coordinates by the attitude data of camera, then the top views of the road surface is got. Experiment result showed that the IPM error caused by the new method was less than 0.8% while the camera attitude was arbitrary. Compared with traditional method, the distortion of road image mosaic is less when the camera is affected by the balleting of the vehicle outdoor. Therefore, the method proposed can not only get highprecision top views of the road surface in the situation of balleting of camera, but also can be applied to the detection of foreign object debris on airport runway and the reconnaissance of damage to the highway.