立体视觉里程计中基于微粒群优化的初始运动估计和内点检测

初始运动估计和内点检测是影响立体视觉里程计定位精度的重要因素.目前,立体视觉里程计都采用基于3点线性运动估计的随机采样一致性(random sample consensus,RANSAC)方法.本文分析了随机采样一致性方法在初始运动估计中的性能:该方法对排除误匹配点是有效的,但在一定采样次数下采样到特征点提取误差和立体匹配误差都很小的匹配点的概率是很小的,所以通过该方法得到的初始运动参数和匹配内点不够精确。本文提出了采用微粒群优化的初始运动估计和内点检测新方法,该方法收敛速度快,搜索精确解的能力强,能够获得高精度的运动参数和匹配内点.立体视觉里程计仿真实验和真实智能车实验表明:和随机采样一致性方法相比,本文方法在运行时间、定位精度方面都更优越.

Initial motion estimation and inliers detection based on particle swarm optimization for stereo visual odometry

Initial motion estimation and inliers detection have an important impact on the accuracy of stereo visual odometry. At present, random sample consensus (RANSAC) method based on the 3-points linear motion estimation is widely used to obtain initial motion parameter and inliers in stereo visual odometry. In this paper, we analyses the performance of RANSAC. It is very effective to eliminate outliers, but the probability which RANSAC have sampled the matching points with low error of feature extraction and error of stereo matching is very low. Therefore, the initial motion parameter and matching inliers computed by RANSAC method are not precise. We propose a new initial motion estimation and inliers detection method based on particle swarm optimization in this paper. Our method has a good performance with the fast convergence and strong global searching ability. Our method can obtain accurate motion parameter and matching inliers. Stereo visual odometry experiments with simulated data and outdoor intelligent vehicle showed that our algorithm outperforms RANSAC method according to run-time, accuracy.