Optimization Of Stero Vision Location Method Based On Point And Line Fearture Matching

In recent years,visual odometry have been widely applied in the fields of robots,augmented reality,virtual reality,unmanned aerial vehicles and self-driving cars.At present,the method of visual odometry based on the point feature is widely used.However,when the environments is low texture,estimating camera motion will fail only with point features,so the low-texture environment will affect the accuracy of the pose estimation.In addition to point features in a structured environment,there are also abundant environmental features,such as points,lines and planes.The level of the line feature is higher than the point feature that it can more,which can more intuitively reflect the geometric information of the surrounding environment.The line feature can be added to compensate for motion estimation failures caused by the absence of point features.Due to the line representation is more complicated than the point,more research space is left in the research of matching algorithm.This paper studies the line matching algorithm and proposes a stereo visual odometry with point and line features.First,for a study on the line matching.On the one hand,the matching function of the line was constructed based on the minimum analysis theory through using the geometric properties of the slope,normal vector and distance,and the method of matching the line with geometric constraints was found.On the other hand,the similarity measure function is constructed by using the gray information of the pixel points,and the line matching method was obtained by constructing the similarity measure function based on the point.Second,for a study of the visual odometry.In terms of the visual odometry with point features relying on the environment,we develop a robust efficient visual odometry.Firstly,we model the stereo visual odometer process which is based on the point and the line feature,when estimating the motion between adjacent images,using the relationship between the quaternion and the rotation matrix to construct the overdetermined equations,and using the least squares solution to solve the overdetermined equations,the initial solution of the model was obtained with Newton iteration.Secondly,we used the visual odometey model to acquire the reprojection error function,and used the Gauss-Newton method to iterative error function obtained the odometry information which was gained after optimizing the initial solution of the modle,which updates the spatial lines by using the Cayley representation with an unconstrained optimization.At last,the odometry information was used to guide the positioning.Finally,we carried out experimental simulation and analysis for the above methods,and used the experimental graphics and numerical data to verify the correctness of our theoretical analysis,and at the same time illustrated the effectiveness of our method.