Research On Depth Perception And Localization Technology Based On Binocular Vision

Distance measurement and positioning technology,as a key technology in computer vision,has important applications in many fields such as autopilot,robot navigation,virtual reality,target tracking,and 3D reconstruction.Binocular stereo vision positioning technology is an important branch of distance measurement and positioning technology.In recent years,deep learning technology has been widely used in industrial production and daily life,and convolutional neural networks have greatly improved the performance of traditional computer vision algorithms in a variety of application environments.However,how to use the convolutional neural network to improve the traditional binocular stereo vision algorithm and obtain accurate positioning results is still a problem to be solved.This thesis summarizes the research status of domestic and foreign positioning technology,and combines it with the convolutional neural network model to apply it to the stereo vision positioning algorithm.The main work done in this thesis is as follows.Firstly,this thesis proposes a stereo matching algorithm based on convolutional neural networks.The algorithm uses a convolutional neural network to fit the matching function,which can avoid manually designing the parameters of the algorithm,reduce the matching error,and improve the performance of the stereo vision algorithm.Secondly,this thesis designs a ranging positioning system based on the matching algorithm described above.Based on the convolutional neural network matching algorithm,this thesis introduces a special coded optical beacon and improves the system processing speed by designing a reasonable beacon coarse positioning algorithm.At the same time,the algorithm converts the disparity map into a three-dimensional point cloud representation.Through clustering and linear regression,the error of the matching algorithm is further reduced,and the positioning accuracy of centimeter level is achieved.Finally,this thesis designs an end-to-end convolutional neural network localization algorithm.The algorithm uses convolutional neural network to realize the steps of stereo matching,cost aggregation and parallax refinement in the traditional stereo vision process.The data-driven method is used to establish the positioning model,and its performance has reached the advanced level in the field.