基于运动线索的单目绝对深度测量

Monocular Absolute Depth Measurement Based on Motion Cues

为实现单目相机识别和测量目标物体的绝对深度信息,提出基于运动线索的测量方法。首先,通过SSD算法和GrabCut算法识别并分割两幅图中的同一目标对象;其次,利用改进的ORB算法和融合几何约束的RANSAC算法获得目标物体的特征匹配结果;再利用Graham Scan算法求得两幅图中特征匹配后的凸包集合,选出最佳匹配的特征线段计算缩放率,最后,通过公式求得目标物体的绝对深度信息。结果表明,当相机移动100 mm时,目标物体测量误差最小为0.94%,最大为5.23%,平均测量误差最小。由此可见,改进的特征提取和匹配算法不仅能均匀化和亚像素化角点,还提高了匹配正确率,同时选出最佳匹配的特征线段也保证了测量精度。

In order to realize monocular camera recognition and measurement of absolute depth information of target objects,a measurement method based on motion clues is proposed.Firstly,the same target object in two images is identified and segmented by SSD algorithm and GrabCut algorithm.Secondly,the improved ORB algorithm and RANSAC algorithm with geometric constraints are used to obtain the feature matching results of the target object.Then the convex hull set after feature matching in the two images is obtained by Graham Scan algorithm,and the best matching feature line segment is selected to calculate the zoom rate.Finally,the absolute depth information of the target object is obtained by formula.The results show that when the camera moves 100 mm,the minimum measurement error is 0.94%,the maximum is 5.23%,and the average measurement error is the minimum.Therefore,the improved feature extraction and matching algorithm can not only homogenize and sub-pixel corners,but also improve the matching accuracy.At the same time,selecting the best matching feature line segment also ensures the measurement accuracy.