改进ORB算法在单目视觉SLAM特征匹配中的应用

针对单目视觉同步定位与地图构建问题对传统定向二进制描述符算法进行改进,结合快速鲁棒特征算法的思想,将尺度空间理论引入传统ORB算法中,同时根据机器人的运动先验信息,预测特征点的可能范围,避免在全局范围内对特征点的检测和匹配.实验表明,改进的ORB算法能显著提高匹配正确率,在多尺度方面表现出色,并能有效减少运算时间,平均耗时14 ms,处理速度约为传统ORB算法的1.3倍、SURF算法的10倍、尺度不变特征变换(Scale-invariant FeatureTransform,SIFT)算法的26倍,适用于单目视觉SLAM问题.

Improved ORB algorithm for monocular vision SLAM

In order to make traditional ORB algorithm suitable for the monocular visual SLAM, this paper proposes an improved ORB algorithm, combines the traditional ORB with SURF by added the scale space to the traditional ORB, and predicts possible position range of feature points according to the movement of the robot to avoid calculation in the global scope. Experiments shows that the improved ORB algorithm can significantly improves the correct matching rate and has good scale invariance. The improved ORB algorithm effectively reduce the operation time, the average time of which is 14ms, and processing speed is about 1.3 times of the traditional ORB, 10 times of SURF, 26 times of SIFT, it is very suitable for the monocular vision SLAM.