融合RANSAC光流跟踪法和特征点匹配法的视觉里程计

为了解决光流跟踪法定位精度不足、误差累积和特征点匹配法耗时久的问题，设计了一种将随机抽样一致性（random sample consensus，RANSAC）光流跟踪法和改进的特征点匹配法结合的视觉里程计。利用RANSAC光流跟踪法对关键帧之间的小规模运动进行估计，RANSAC算法对光流跟踪的误匹配点进行剔除，大大降低了光流跟踪法存在的误匹配；而关键帧之间的运动估计则利用改进的特征点匹配法，以修正光流跟踪法的估计误差；最后利用卡尔曼滤波将RANSAC光流跟踪法和改进的特征点匹配法进行融合。实验结果表明:该文的算法能够克服光流跟踪法精度不足、误差累积的问题，将平均相对误差由15.5%提升到了2.6%；同时也能在一定程度上提高特征点匹配法的速度，将特征点匹配法的平均耗费时间由37.28 ms提升到了21.07 ms。

Design of visual odometer based on RANSAC optical flow method and feature point matching method

In order to solve the problem of insufficient positioning accuracy, error accumulation and long time consuming of feature point matching method, a visual odometer that combines RANSAC optical flow method and improved feature point matching method was designed. The RANSAC optical flow method was used to estimate the small-scale motion between key frames. The RANSAC algorithm eliminated the mismatch points of optical flow, which greatly reduced the mismatch existing in the optical flow method. The motion estimation between key frames used the improved feature point matching method to correct the estimation error of the optical flow method. Finally, the RANSAC optical flow method and the improved feature point matching method were fused by using the Kalman filtering. The experimental results show that the algorithm can overcome the problems of insufficient accuracy and error accumulation of the optical flow method, which increases the MRE from 15.5% to 2.6%. And it can also improve the speed of the feature point matching method, which increases the average consuming time from 37.28 ms to 21.07 ms.