基于双目相机的视觉里程计

Visual odometry based on binocular camera

针对移动机器人视觉导航定位需求,提出一种基于双目相机的视觉里程计改进方案。对于特征信息冗余问题,改进ORB(oriented FAST and rotated BRIEF)算法,引入多阈值FAST图像分割思想,为使误匹配尽可能减少,主要运用快速最近邻和随机采样一致性算法;一般而言,运用的算法主要是立体匹配算法,此算法的特征主要指灰度,对此算法做出改进,运用一种新型的双目视差算法,此算法主要以描述子为特征,据此恢复特征点深度;为使所得位姿坐标具有相对较高的准确度,构造一种特定的最小二乘问题,使其提供初值,以相应的特征点三维坐标为基础,基于有效方式对相机运动进行估计。根据数据集的实验结果可知,所提双目视觉里程具有相对而言较好的精度及较高的实时性。

Aiming at the visual navigation positioning requirements of mobile robot,a visual odometry based on binocular camera was proposed.For the problem of feature information redundancy,the improved ORB(oriented FAST and rotated BRIEF)algorithm was introduced,and the multi-threshold FAST image segmentation idea was introduced.To reduce mis-matching as much as possible,the fast nearest neighbor and random sampling consistency algorithms were mainly used.Generally speaking,the algorithm used is mainly a stereo matching algorithm.The characteristics of this algorithm mainly refer to grayscale.This algorithm was improved and a new binocular disparity algorithm was used.This algorithm mainly used descriptors as features.Therefore,the feature point depth was restored.To obtain the relatively high accuracy of the pose coordinates,a specific least squares problem was constructed to provide an initial value,and based on the corresponding 3D coordinates of the feature points,camera motion was estimated based on efficient way.According to the experimental results of the data set,the binocular vision mileage of this design has relatively good accuracy and high real-time performance.