移动机器人视觉图像特征提取与匹配算法

Robust image feature extracting and matching algorithm for mobile robots vision

针对移动机器人导航过程中视觉图像处理速度慢以及特征点提取与匹配实时性和准确性差等特点,提出了一种基于SIFT特征提取算法与KD树搜索匹配算法相结合的新方法,通过对候选特征点进行多次模糊处理,使其分布在高斯差分图像的灰度轮廓线边缘,利用SIFT特征提取算法找到满足极限约束的极值点;通过KD树最邻近点搜索和匹配算法使处理后的特征点与原始图像进行特征匹配,快速找出匹配正确的特征点。实验证明,该方法对环境光照、视野角度频繁变化的环境具有较强的鲁棒性,能够满足移动机器人自主导航过程中对视频图像处理的实时性和准确性要求。

According to the real-time and accuracy requirement to process image during navigation for mobile robots, this paper proposed a new feature extracting algorithm, SIFT algorithm, which combined the searching and matching algorithm, KD tree. Firstly, fuzzed the feature images many times, so that distributed those extracted features around gray outline of Gaussian difference image. Then the global Best points, which satisfied extreme constraints, were obtained based on SIFT algorithm. Finally, the exact matching features could be found by using matching algorithm based on KD tree quickly. It is validated that the algorithm is strongly robust to the environment change from the experiments, such as illumination and angle of view, and is able to meet the requirements of real-time and veracity in the process of navigation for mobile robots.