Pose Estimation Of Mobile Robot Based On Monocular Vision

Pose estimation of visual-based mobile robots refers to estimate the location and posture of mobile robots with the help of monocular cameras.The traditional sensor cannot avoid the problems in the location aspect,the odometer is susceptible to the wheel slip,and GPS is vulnerable to the obstruction and electromagnetic interference.In addition,the detected distance of the infrared sensor is limit.This study can solve the above problems by using vision sensors,and improve the adaptability of the research of mobile robots' pose estimation.Thus,mobile robots have a very broad application prospect in the fields of aerospace,deep sea exploration,military and so on.Monocular vision has been widely used because of its' low cost,and moreover,it does not need to solve the problem of stereo matching.But there are some new problems,the real-time pose estimation cannot be guaranteed by the inefficient feature matching and how to solve the ambiguity of the results for existing pose estimation algorithms.In addition,most pose estimation algorithms can only apply to the plane scene or stereo scene.Hence,two image-based pose estimation techniques are proposed in the thesis to solve the above problems.In general,the main work in the thesis is summarized as follows:1.For the low efficiency of SIFT feature matching algorithm,a feature matching algorithm of PCA-SIFT is proposed.The algorithm can not be affected by rotation and illumination,and it can reduce computational complexity to improve real-time of the whole mobile robot vision system.2.For the problem of high real-time requirement of indoor pose estimation,combined with the motion constrain in two dimension plane,a fast pose estimation algorithm based on homography is adopted.This method can effectively solve the ambiguity problem of pose solution.The experimental result shows that the present method can obtain the position and pose of the mobile robot accurately.3.In view of some scenes that features do not exist in the same plane,a pose estimation method is applicable for both planar scenes and stereo scenes is proposed.Comparing with traditional SFM methods based on factorization,the proposed method can obtain the correct constraint matrix and get results with higher accuracy.Experimental results demonstrate that the proposed method can realize the right pose estimation of mobile robot and 3-D structure recovery.Finally,the conclusion is provided and the possible further works are presented.