Research On Visual Odometer Based On Improved Point-line Feature

Vision odometer is mainly used to solve the positioning problem of mobile robot.When the mobile robot is in an unknown environment,it can use the image information collected by its own sensor to judge its position.Nowadays,there are still many problems in visual mileage calculation.The main point feature algorithm does not perform well in the scene with missing texture and obvious illumination changes,which affects the accuracy of motion estimation.In order to improve the above problems,this paper studies the visual mileage calculation method,and proposes a more robust visual mileage calculation method by introducing the improved line feature matching algorithm and integrating the advantages of point and line features on the premise of real-time.The main contents of this paper are as follows:(1)Based on the analysis and comparison of the current mainstream feature detection algorithms,the orb feature extraction algorithm is finally selected and improved on the basis of the original algorithm combined with the ROI search domain algorithm.By dividing the detection area,under the premise of ensuring the quality of point features,the calculation cost is saved.(2)In view of the influence of light changes and the lack of point features in the under texture environment,the cumulative error of traditional visual odometer will increase with the displacement of robot,which leads to the inaccurate estimation of pose,a strategy of fusing line features is proposed.Firstly,the line features are extracted by LSD algorithm,and then the line matching process is optimized according to the geometric relationship between the line features,which is compared with the traditional computational description algorithm Compared with the algorithm,it can greatly improve the matching speed and the robustness of the algorithm to the illumination changing environment.(3)The camera motion between keyframes is optimized by building a graph model based on the point line feature,and the improved keyframe selection algorithm is used to limit the distance between frames,to ensure that the distance betweenframes is within a certain range,to avoid the influence of redundant data on the operation efficiency of the system,and to match the line matching algorithm used in this paper.Finally,a visual odometer system based on Kinect depth camera is built to verify the performance of the algorithm.Through the test in real scene and multiple different data sets,it is proved that the algorithm proposed in this paper can meet the real-time requirements,and has better positioning accuracy in the scenes with obvious under texture and light transformation.