基于改进SIFT算法的双目视觉距离测量

针对视觉传感器距离测量中所使用的图像特征匹配算法精度不高、计算量大、实时性差等问题,提出了一种改进尺度不变特征变换(SIFT)图像特征匹配算法,并应用于双目测距系统当中.改进SIFT算法基于简化尺度构造空间,以曼哈顿距离作为最邻近特征点查询中的相似性度量,提高了算法效率.初次匹配之后与随机采样一致算法(RANSAC)结合,剔除误匹配点;基于精度较高的二次匹配点,提取匹配点像素信息进行距离计算,通过测距试验验证算法的可行性.实验结果表明:提出的方法获取目标距离达到较高精度,满足观测设备要求.

Binocular stereo distance measurement based on improved SIFT algorithm

Aiming at problems such as low precision,large amount of calculation and poor real-time performance of image feature matching algorithm that used in distance measurement by visual sensor,an improved scale invariant feature transform (SIFT ) image feature matching algorithm is proposed and applied to the binocular distance measurement system. In order to improve the efficiency of the algorithm,the improved SIFT algorithm is based on the simplified scale structure space,which takes Manhattan distance as the similarity measurement in the nearest neighbor query. To eliminate mismatching points,the random sample consensus(RANSAC)algorithm is used after first match. Based on the second match points with high precision,the pixel information of match points is extracted to calculate distance,and the feasibility of the algorithm is verified by distance measure experiment. Experimental results show that the distance of target with high precision meets the requirements of observation equipment based on the proposed method.