Research on multi-resolution image matching method based on binocular stereo vision
In binocular stereo vision systems,noise can damage image features and increase the difficulty of image feature extraction in complex scenes,leading to a decrease in matching accuracy and robustness.Therefore,a multi-resolution image matching method based on binocular stereo vision is proposed.This method aims to effectively obtain information from images of different scales and achieve high-precision matching.In this method,binocular stereo vision model with a binocular rotating camera is utilized to scan and image the objects,and internal and external space calibration is used to improve the positional accuracy of the binocular rotating camera and ensure the multi-resolution imaging effect of the objects.The imaging effect is input into the pyramid stereo matching network,and the binocular image features are extracted by the pyramid-like multiple atrous convolution operation in the network.And then,their texture feature details are enhanced based on variable convolution.Finally,the binocular image matching is achieved in combination with fine-grained features and mutual attention mechanism.The test results show that after spatial calibration,the minimum imaging errors of the left and right cameras are 0.6 Pixel and 0.4 Pixel,respectively;both of the mean value and the variance of the coordinates deviation of the matching points are lower than 0.012 and 0.011,respectively,and the matching effect is good.
万方数据
维普
