摘要
研究双目立体视觉技术,特征的提取和匹配是双目视觉的最基本的问题。目前,SIFT已经被证明鲁棒性最好的局部不变特征描述符。但是SIFT算法产生的误匹配较多,精度偏低,为了解决这一问题,同时降低算法特征提取与匹配的复杂度,达到双目立体视觉实时性的要求,文中提出了一种结合小波变换和SIFT特征点的双目立体视觉匹配方法。首先,对双目视觉系统采集的左、右图像进行小波分解,把分解得到的低频图像作为输入,用SIFT算法进行特征点的初始匹配,再利用极线约束的理论求得精确匹配。实验结果表明,该方法具有较强的适应性,能够在减少误匹配的同时,大大加快运算速度。
The feature extraction and matching is the basic problem for binocular stereo. SIFT has proved to be the most robust local invar iant feature descriptor in object recognition and matching. Howerver, it generates the mismatch and low precision. In order to solve this problem and reduce complexity of the SIFT algorithm, meeting the higher accuracy and real-time requirements of binocular stereo vision, present an algorithm of binocular stereo vision matching approach based on wavelet transformation and scale invariance feature transfor- mation { SIFT ）. First, process the left and fight images respectively collected by binocular stereo vision system based on the method of wavelet decomposition. Then complete the initial match of feature point to low frequency image obtained by decomposition with SIFT al- gorithm. Finally, achieve exact match in accordance with epipolar constraint theory. The experimental results prove that this algorithm has a strong adaptability. It makes high matching accuracy and enhances the computing speed.
出处
《计算机技术与发展》
2012年第11期81-84,共4页
Computer Technology and Development
基金
广东省教育部产学研结合项目(2010B090400186)
