基于SIFT特征的月面模拟环境视差估计

Disparity Estimation for Simulated Lunar Environment based on SIFT Features

根据对月球车双目立体视觉系统及其工作环境特点的分析,给出了一种基于SIFT特征的月面模拟环境视差估计方法;首先通过尺度不变特征变换方法从图像中提取关键点作为特征点,然后以特征点特征向量的相异度作为匹配度量进行特征点匹配,最后通过已匹配特征点的种子扩散来进行视差的估计;与传统的SIFT特征匹配算法相比,通过减少特征点的维数可以提高月面模拟环境图像的匹配速度,而利用行扩散的传播方式可以提高匹配精度;实验结果表明,该算法用于室外地形图对的匹配时,可以生成较为准确和稠密的视差图,证明了该算法的实用性和有效性。

A disparity estimation algorithm for simulated Lunar environment based on SIFT features is brought forward based on the analysis of the stereo vision system and working surroundings of Lunar Rover. Firstly, the keypoint was extracted by scale--invariant feature transform method from image as feature points. Then the feature points were matched by comparing dissimilarity of keypoint deseriptor; Lastly, when the matched features points were belief propagated in the whole image, the disparity of the image had been estimated. The pro-posed algorithm could improve matching computing speed and accuracy because of reducing the number of dimensions and adopting line--growth method compared to traditional feature--matching algorithms. Experimental results indicate that the dense disparity map could be generated with high accuracy when used in the matching of stereo image pair of outdoor terrain and show the practicality and validity of the proposed algorithm.