基于角点聚类的移动机器人自然路标检测与识别

Corner clustering based detection and recognition of natural landmark for mobile robot

针对未知环境中机器人视觉导航的自然路标检测,提出了一种基于角点聚类的自然路标局部特征提取、不变性表示及其匹配算法.用SUSAN算子提取左右视图中的角点,在极线约束下对左右视图的角点进行匹配,消除遮挡或噪声引起的角点;同时应用立体视觉计算角点视差,进一步筛选角点.根据角点聚类策略提取自然路标局部特征,并提出不随距离、角度变化的局部特征不变性表示及匹配方法.理论分析和实验结果表明,该算法具有较好的鲁棒性,在一定距离和角度变换下能够对路标进行正确识别.

Aiming at the detection of the natural landmarks for mobile robot navigation based on vision system in unknown environment, this paper presented a novel method based on corner clustering method to build the invariant model of local feature of unknown natural landmark and recognize it. Firstly, SUSAN operator was used to detect corners in the left and ri two pictures, so that occluded points could be delete ght pictures. Then those corners were matched in the d. Then, corners were clustered according to the eorner clustering strategy to build the invariant model of local features of unknown natural landmark. Lastly, the method was given to match the model, and experiments showed that the method was effective to build the invariant model of natural landmark and could also be recognized when viewpoint and scale were changed.