Research On Stereo Matching Algorithm For Real-time Edge Sensing

Stereo matching has always been a research hot spot in computer vision,which mainly matches parallax by calculating the cost of two digital images.Among them,binocular stereo matching has attracted wide attention of researchers because of its bionic characteristics.Binocular stereo matching data sets use two cameras to obtain two digital images of three-dimensional scenes from different time and space.This technology is widely used in unmanned aerial vehicles,virtual imaging and other fields.This paper studies the real-time edge-aware stereo matching algorithm.Firstly,the background meaning and edge-aware matching algorithm of traditional stereo matching are introduced,then Hilbert stereo matching algorithm and real-time binocular stereo matching fusion network at different spatial levels are studied,and then the thermodynamic color guidance mechanism is introduced as the supervision factor of the whole network.Finally,the effectiveness and efficiency of real-time edge-aware stereo matching algorithm are verified and illustrated by training and testing on large stereo matching benchmark data sets.The main research contents include:1.In the calculation and aggregation of disparity cost,a disparity extraction method based on Hilbert transform with end-to-end structure design is proposed.Different from the traditional method which acts on parallax itself,this method analyzes and obtains the cost information of the image through Hilbert filtering,which is used for the following operations such as parallax matching calculation,thinning and enhancement.2.In disparity matching and thinning,an innovative multi-resolution pixel disparity reconstruction and thinning design is proposed,and a stereo matching algorithm based on multiple Hilbert is proposed.It is worth noting that in the disparity refinement step,the Hilbert stereo matching network is mainly composed of several different levels of predicted Hilbert coefficients,so that the purpose of this operation is to obtain enough disparity features,and to some extent,it solves the problem that disparity regression cannot be performed when the image is blocked and reflected.3.Based on the research of Hill disparity extraction and multiple Hilbert stereo matching algorithms,this paper proposes a real-time edge-aware stereo matching algorithm,which refines the disparity in different frequency domains,suppresses the edge phase error,and enhances the final output image refinement effect.At the same time,the thermodynamic color guiding factor is adopted to speed up the learning of disparity in the network,and the overall fusion end-to-end architecture is adopted to improve the speed.Through a large number of experiments,it is proved that this operation can also greatly improve the disparity matching accuracy of the algorithm and accelerate the running speed of the algorithm.