Research On Application Technologies Of Binocular Stereo Vision Based On Polarization Imaging

Binocular stereo vision technology is widely used in industrial precision measurement,automatic driving,robot autonomous navigation,medical diagnosis and other fields because of its non-contact,simple structure and other excellent characteristics.At the same time,the traditional binocular stereo vision technology is also faced with the problem of information perception of transparent target,low texture target,high reflective target,plane image and other special targets.In order to deal with the above problems better,this dissertation studies the application of polarization imaging in binocular stereo vision based on the construction of polarization binocular stereo vision system,characteristics and evaluation of polarization imaging image,multi polarization information fusion algorithm,image stereo matching algorithm,3D visualization and analysis of spatial point set and other key technologies,It has important theoretical innovation and wide application prospects to enhance the target perception ability of traditional binocular stereo vision system and the ability to deal with special targets.The main achievements and innovations of this dissertation are as follows:1)The binocular stereo vision system based on polarization camera is designed and constructed,and the output image characteristics of polarization camera are analyzed.Polarization imaging is introduced into binocular stereo vision,which provides a new solution to the problem of 3D information acquisition of special target and recognition of plane image target.2)Aiming at the distortion of objective index evaluation under the influence of noise,two composite evaluation indexes,unit information entropy and unit average gradient,are proposed.The polarization imaging effect of transparent target,low texture target,high reflection target and plane image target is evaluated by using objective evaluation index.Compared with the visible light image,the average gradient of polarization image is 9.16 times higher,the unit average gradient is 12.75 times higher,the information entropy is 1.02 times higher,and the unit information entropy is 2.04 times higher3)The multi polarization information image fusion algorithm based on NSST and morphological gated polarization image fusion algorithm based on mean filter decomposition are proposed.The multi polarization information gated fusion algorithm based on NSST achieves high-quality fusion of multi polarization information through fine fusion regulation of image components by gating the size of regional variance.The average gradient,unit average gradient,information entropy and unit information entropy of fused image are 1.99 times,6.95 times,1.09 times and2.11 times higher than those of visible image.The morphological gated polarization image fusion algorithm based on mean filter decomposition is gated by the gray threshold after morphological operation,and the algorithm is efficient.The average gradient,unit average gradient,information entropy and unit information entropy of the fused image are 2.05 times,6.77 times,1.06 times and 1.91 times higher than those of the visible image.The two fusion algorithms can maintain the important feature information of the three source images,suppress the image noise,and achieve the high-quality expression of the fused image.4)Analyze and summarize the stereo matching strategies for fused images and low-quality images,and propose a stereo matching algorithm based on image restoration.The stereo matching algorithm based on image restoration adds image reverse restoration branch network to pyramid stereo matching neural network algorithm,constructs multi branch neural network structure,reduces the influence of noise and low contrast on image feature extraction,and solves the stereo matching problem of polarization fusion image and low quality image.The visualization analysis software of spatial point set is designed and developed to realize the visualization analysis of 3D point set,which provides convenience for better use of3 D data.