大视场双目视觉定位系统中多目标稀疏匹配

Multiple targets sparse matching for binocular vision positioning system with large field of view

野外远距离、大视场、双目视觉物体定位监测系统是基于双目立体视觉原理的三维坐标测量系统。存在随机出现的多个无特征目标的正确匹配率低,导致目标定位精度降低的共性问题。正确判断首个目标出现,对于同一目标在同侧图像序列连续出现,形态特征由大到小变化或分裂的情况,利用质心判断方法排除伪目标,检测目标正确质心位置;然后,两段图像序列时间同步,同一时刻多个目标基于极线约束进行稀疏匹配,获得同一目标在两台相机中正确的对应关系。实验证明,现场连续采集的9帧图像中包含8个目标,利用上述方法完成目标检测及匹配,提高了准确率。

A positioning and monitoring system for the objects with long distance, large view field and binocular vision was a three-dimensional coordinate measuring system based on binocular stereo vision principle. There was a common problem that the low correct matching rate of multiple non-feature targets randomly appearing led to the reduced accuracy of target positioning. The key was judging the trigger start time correctly. For the situation of the same target appearing continuously in the same side of the image sequence, or the morphological characteristic of the target changing from large to small or splits, the use of centroid method was to eliminate pseudo targets and detect target centroid position correctly. Then image sequences had time synchronization, and multiple targets realized sparse matching based on polar constraint at the same time. The experiment proves that 9 frame live continuous images contain 8 targets, using the above method to complete the targets detection and automatic positioning, and improve the accuracy rate.