基于光束法平差的视觉里程计算法研究

Study on visual odometry algorithm based on bundle adjustment

视觉里程计是通过单个相机或多个相机所获得的图像序列作为输入来估计自身运动的方法。应用领域涵盖机器人、可穿戴计算、增强现实和自动驾驶等。本文进行了随机抽样一致性算法和利用Huber核函数约束的光束法平差算法的研究,并在公开的KITTI标准数据集上进行了算法实验验证。实验结果表明:随机抽样一致性算法可以实时有效地进行异值点剔除。利用Huber核函数约束的光束法平差后,在100m的路径上的视觉里程计平面位置误差RMS值为2.09m,平面位置精度提升约17.72%。

Visual odometry(VO)is the process of estimating the egomotion of an agent(e.g.vehicle,human,and robot)using only the input of a single or multiple cameras attached to it,and it is widely applied in robotics,wearable computing,augmented reality,automotive,etc.In this paper,RANSAC algorithm and bundle adjustment with Huber function are discussed,and these algorithms are tested based on KITTI dataset.The results show that RANSAC algorithm can efficiently improve the performance of outlier rejection in real time,and the RMS of the visual odometry horizontal position error is 2.09 m on a 100-m path,,which is improved by 17.72%after using bundle adjustment with Huber function.