Deep Learning Based Binocular Vision And Structured Light Technology

Stereoscopic vision is an important research topic in the field of computer vision.Its purpose is to reconstruct the three-dimensional information of the scene and make the machine have visual perception ability.Stereo vision technology has been applied in all walks of life,such as robot navigation,driverless operation and so on.Among these stereoscopic vision technologies,binocular stereoscopic vision technology and structured light technology are most commonly used.Binocular stereo vision technology uses two cameras to simultaneously acquire two images of the object to be measured,and obtains three-dimensional information of the object by calculating the positional deviation between the corresponding points of the image.The structured light technology uses the projected modulated pattern to the object to be measured,acquires an image of the measured object through the camera,and then obtains depth information based on the triangulation principle..At present,the field of two-dimensional vision is dominated by deep learning.Although the field of stereo vision is still based on geometry-based methods,many excellent stereo vision techniques based on deep learning have emerged.Therefore,this paper combs the current deep learning based binocular stereo vision technology and structured light technology,and investigates the current excellent articles to provide guidance for subsequent research.The dissertation first introduces the basic principle of binocular stereo vision method and structured light method.Then the dissertation focuses on the important modules of deep learning based binocular stereo matching technology,including feature extraction,feature matching,cost aggregation,disparity prediction and its refinement.For each network module,the dissertation reviews in detail the processing methods of each article in recent years,finds the similarities and differences between them in processing,analyzes their internal principles,and evaluates their improvements to each module.Subsequently,the dissertation introduced two methods of structured light technology,one is structured light technology based on random forest method,and the other is structured light technology based on deep learning.