基于掩模和俯仰角补偿的视觉自运动估计  

Monocular Egomotion Estimation of Mobile Robot Based on Masked M-Estimator and Pitch-Compensated Algorithms

作　　者：王彩玲[1] 赵春霞[2] 

机构地区：[1]南京邮电大学自动化学院,南京210003 [2]南京理工大学计算机科学与技术学院,南京210094

出　　处：《系统仿真学报》2012年第11期2319-2323,2328,共6页Journal of System Simulation

基　　金：国家自然科学基金重大研究计划重点项目(90820306);国家自然科学基金重点项目(60632050);国防基础技术研究项目(K1702020302)

摘　　要：提出了一种基于掩模M估计和俯仰角补偿的移动机器人自运动的单目视觉估计方法。在6-DOF自运动模型的基础上,提出了一种适合于乡村道路环境的4-DOF自运动估计模型,以此为基础,采用一种多尺度的超复小波相位相干理论来鲁棒地估计缺乏规则纹理的路面图像的微小运动,设计了一种掩模M估计对运动向量进行优化估计,并提出了一种俯仰角补偿算法克服乡村路面颠簸有效地估计机器人偏航角、俯仰角增量和平移增量。在乡村道路环境下,利用某自主机器人采集的实际数据对算法进行了仿真试验,运动估计结果与车载GPS/INS捷联系统做了对照。实验结果表明了自运动估计算法的有效性。Egomotion estimation based on masked M-estimator and pitch-compensated algorithms were proposed. A 4-DOF egomotion estimation model was designed for the mobile robot on country road, the image motion was estimated through an efficient hyper-complex wavelet phase coherent theory, which could robustly detect the multiscale image motions even if there are few regular textures in country road. Pitch-compensated and masked M-estimator algorithms were proposed to estimate the instantaneous steering angle, pitch angle and traveling distance efficiently. The algorithms were successfully validated on videos from an automotive platform on the country road, and the egomotion estimation results from our method were compared to the ground truth measured with a GPS/INS joint system. These experimental results indicate that the algorithms are effective.

关 键 词：单目视觉 自运动估计 俯仰角补偿 掩模M估计 

分 类 号：V448.22[航空宇航科学与技术—飞行器设计]