Research On Stereo Matching Algorithm Based On Binocular Vision

Binocular stereo vision occupies an important position in the field of artificial intelligence and is an important branch of computer vision research.It is widely used in robot navigation,video surveillance,medical diagnosis,military reconnaissance,and unmanned driving.Binocular stereo matching simulates the principle that human eyes see things and processes information in the brain.The pixel matching method is used to determine the correspondence between the pixels in the left and right binocular images obtained by the binocular camera and calculate the depth information in three-dimensional space.Finally,the image of the object is restored to a three dimensional scene through information such as the angle,color,and edge features of the image,so as to achieve t he reverse modeling effect from the plane to the three-dimensional.This paper from two aspects to mainly researche the algorithm of real-time binocular stereo matching: designing and recursive implementation.On the one hand,it researches and designs binocular stereo matching algorithms with lower algorithm complexity and higher matching quality;The proposed algorithm is implemented recursively to achieve the goal of effectively solving the binocular stereo matching real-time.The main research contents are as follows:First,this article studies the basic theory of binocular stereo matching,compares and analyzes different stereo matching methods,summarizes the advantages and disadvantages of various algorithms,and identifies areas for improvement.Then,based on the bilateral filtering adaptive weighting algorithm,by adding a boundary strength term,a new three-sided filtering adaptive weighting binocular stereo matching algorithm is proposed,and the boundary between adjacent pixels is calculated by the local energy model strength to improve matching accuracy.Inspired by the high matching rate of recursive filtering,the recursive edge-preserving filter technology was introduced into the cost aggregation process,making it an aggregation method with constant algorithm complexity,greatly reducing the running time of the matching algorithm,greatly The speed of stereo matching is improved,and the real-time requirement of binocular stereo matching is met.Finally,experiments are performed on the Midd lebury benchmark test set and compared with other stereo matching algorithms.Through quantitative evaluation from the aspects of matching accuracy,matching efficiency,algorithm complexity,etc.,the experimental results show that the average mismatch ra te of the proposed algorithm is 4.91 %,The matching accuracy is higher than the same type of binocular stereo matching algorithm,with an average matching speed of 258 ms,which meets the real-time requirements of binocular stereo matching.This paper strictly follows the steps of binocular stereo matching,from camera imaging models to final parallax optimization,and conducts in-depth research on real-time binocular stereo matching.Finally,an adaptive weighted stereo matching algorithm with recursive trilateral filtering is proposed to improve stereo matching accuracy and speed to meet the requirements of binocular vision accuracy and real-time performance.The research results of this paper have broad application prospects in the field of real-time binocular vision based on the universal PC platform.