Research On Image Feature Points Matching Algorithm Of Intelligent Vehicle Vision Odometer

With the development of computer vision,more and more visual sensors are used to realize the positioning and motion estimation of intelligent vehicles.In order to realize the autonomous navigation of the intelligent vehicle,the vehicle must be able to achieve self-positioning during the movement so as to obtain the current position and posture information.The visual odometer technology for obtaining the information of the vehicle body itself through the image sequence obtained by the on-board camera can be used as a tradition.The important addition of the odometer also lays the foundation for visual orientation and map construction(SLAM).The purpose of this paper is to conduct an in-depth study of image feature point matching algorithm,a key technology in the visual odometer of intelligent vehicle navigation systems.The task of image matching is to establish the correspondence between the same scene parts in the two images.SIFT algorithm has a milestone significance as a local feature image matching method and shows good performance in image matching process.However,because the SIFT algorithm has a large amount of computation,the matching process takes a lot of time,and it is not suitable for applications with high real-time requirements.The binary-based feature descriptor methods proposed in recent years,such as the BRIEF and ORB algorithms,have the characteristics of low computational cost and small memory usage,which promote the development of feature point image matching,but such algorithms lack support for scale invariance.Feature points are susceptible to image scale changes in the environment,viewing angles,and noise.Based on previous researches,this dissertation deeply studies the feature points extraction and matching algorithms in visual odometry,and proposes a new image feature point matching algorithm.The main work of this paper is as follows:1.A feature extraction algorithm based on the scale space is proposed.The Gaussian Laplacian operator is used to construct a scale space using a center-wrapped double-layer filter,and each pixel in the original image is calculated in a multi-scale space.The center surrounds the Haar wavelet response value,uses the integral image to accelerate the entire operation process,uses the non-maximum suppression method to detect the extreme value,and finally obtains more stable feature points through Harris and sub-pixel interpolation.2.In view of the fact that the binary feature descriptors are susceptible to noise in the environment during the matching process,in order to ensure the matching accuracy of the feature points,this paper proposes an image matching and purification method based on bidirectional pre-filtering based on the RANSAC algorithm.Before purification,a high-quality sample set was filtered through bi-directional matching.Then,the sample was purified and optimized using the RANSAC algorithm,and the false matching points were eliminated.Experimental results show that the algorithm can achieve high matching accuracy and efficient calculation efficiency.A visual distance calculation method based on image feature point matching is proposed.Firstly,the image feature point matching algorithm proposed in this paper and the improved RANSCA algorithm are used to obtain the correct pair of matching points in two adjacent frames.The singular value decomposition method is used to solve the algorithm.The initial movement parameters of the car,and the use of maximum likelihood estimation to further optimize the accurate movement parameters,and finally proved by experiment that the visual odometer has a good accuracy.