Research On Monocular Vision Odometry Calculation Based On Feature Point Method

Visual odometry is one of the methods of visual positioning.Video information is obtained by single or more cameras equipped on vehicles,robots and other carriers,and then the motion of the carrier is estimated by analyzing the continuous frame image.In view of the shortcomings of traditional global positioning system and inertial navigation system,such as high cost and limited use scene,visual odometer has the advantages of low cost,small cumulative error and wider use of scene.It plays an important role in field operation,indoor navigation,underwater navigation,aerospace and other fields,which can cooperate with the traditional positioning methods and effectively supply the positioning information.This paper studies the visual odometer,and proposes a monocular visual odometer calculation method based on the corner feature extraction of the fusion pyramid and shi-tomasi.Considering that compared with stereo cameras or 3D cameras,this method uses a monocular camera as an image sensor because of its low cost,small amount of calculation,and simple algorithm.Considering that the visual odometer method based on feature points is not sensitive to external factors such as lighting and the algorithm is stable,the algorithm mode based on feature point is selected.This method addresses the shortcomings of the traditional visual odometer calculation method in the process of corner extraction,the distinguishability and repeatability of feature points,and the disadvantages of low positioning accuracy and large pose information errors.A new method of corner extraction based on Gaussian pyramid and shi-tomasi is proposed.The algorithm uses the pyramid LK optical flow method to complete the corner feature tracking,and finally uses the eight-point method to calculate the motion information.The thesis first introduces the research status of visual odometer in detail,and then briefly introduces the basic principles of visual odometer calculation method.Based on this,the detailed description of corner extraction algorithm based on image pyramid,pyramid LK optical flow tracking algorithm,and eight-point Method of motion recovery algorithm.In this paper,the KITTI dataset is used as the experimental object,and the method in this paper is compared with the traditional visual mileage calculation method using the three corner extraction methods of shi-tomasi,fast and Harris.The experimental results show that compared with traditional algorithms,the method in this paper can effectively improve the clustering of corner points,improve the distinguishability and repeatability of feature points,effectively improve the positioning accuracy of the visual odometer,and enhance the algorithm for different environments Adaptability.