针对结构光双目三维重建方法中存在误匹配点较多的问题, 提出了一种基于相机抖动的多方位误点淘汰筛选算法。利用平面镜轻微移动模拟相机抖动, 对复原图像进行重新投影, 进行误点的淘汰筛选, 提高复原精度。实验结果表明, 该方法与未抖动相机时比较, 单幅图复原误点明显减少, 多幅图平均精度提高20%~30%, 精度平均水平达到0.16 mm, 可以实现对机车轨道和轮对检测的高精度三维测量, 满足实际检测的高精度要求。

Concerning large numbers of mismatched points in the binocular 3d reconstruction methods based on structured light, ascreening and eliminating algorithm of multi-directional mismatched points based on cameras’ jittering is discussed. The restorated image is reprojected by using a plane mirror slightly moving to simulate camera jittering to eliminate and screen the mismatched points so that the restoration accuracy is improved. The experiment result indicates that the number of mismatched point is significantly reduced comparing withthat of no jittering algorithm, and the average accuracy of multiple pictures improves 20%~30%, and the average accuracy reaches 0.16 mm, which can realize high-precision 3D measurement of locomotive track and wheel pair detection. The proposed method can satisfy the practical testing requirements.