针对传统图像识别视角单一问题,提出了一种改进的多视角目标识别算法,结合空间结构连续性,利用迭代最邻近算法ICP(Iterative Closest Points)进行多视角目标识别.首先,构建高斯尺度空间来提取具有尺度不变性的SURF(Speed-up Robust Features)特征点,得到特征描述子,且根据描述子的相似性得到初始转换参数;然后,将描述子相似性与空间结构连续性相结合,利用ICP求得匹配对,并进行配准;最后,在各个角度的图像中对目标图像进行识别,取得了较好的效果.实验结果表明,所提出算法识别效果显著,准确率高,具有很好的鲁棒性.%Object recognition plays a great role in machine vision.Traditional image recognition just considers only viewpoint.An object recognition system has been developed in this paper.Firstly,Gaussian scale space is built to extract scale invariant feature points, and the SURF descriptor is utilized to characterize those feature points and get the transformation parameters.Then, feature similarity is combined with spatial consistency based on the iterate closest point.Finally, object is recognized in multiple view.Comparative experiments show that the algorithm has high performance in accuracy and robustness.
展开▼