煤矿救援机器人环境目标图像识别

Environmental target image recognition of coal mine rescue robot

针对标准尺度不变特征变换(SIFT)算法存在搜索视觉图像中关键点出现计算冗余和目标识别实时性差的问题,提出了一种改进的SIFT算法,并将其应用到煤矿救援机器人的环境信息感知和目标识别匹配中。该方法以马氏距离代替标准SIFT算法中的欧氏距离,简化了特征点提取,避免了特征点的误匹配。现场试验结果表明,改进后的SIFT算法提高了煤矿救援机器人对煤矿井下环境目标识别的实时性和目标匹配的准确性,为煤矿救援自主移动机器人实现避障、行走做好了视觉前提。

In view of problems that traditional scale-invariant feature transform（SIFT）algorithm has calculation redundancy in searching key points of vision images and poor real-time performance in target recognition,an improved SIFT algorithm was proposed which was applied to coal mine rescue robot to realize environmental information perception and target recognition matching. The improved SIFT algorithm adopts Mahalanobis distance to replace Euclidean distance in the traditional SIFT algorithm and simplifies extraction of the image feature points,avoids mismatching of feature point.The field test results show that the improved SIFT algorithm improves real-time performance in underground environmental target recognition and accuracy of target matching of coal mine rescue robot,which makes visual premise for realizing obstacle-avoidance and walking of coal mine rescue autonomous robot.