Research Of Binocular Vision Algorithm And Design The System Based On DSP

The binocular stereo vision is one of the basically classical issues, and also one important form of the machine vision technology. It is more and more widely used in the Robot Vision, Autonomous Driving, Human Flow Statistic and so on. The binocular stereo vision is the technology of imitating human's eyes using machine by people to apperceive and recognize the 3D scenes and subjects, through a pair of images from the two different viewports.This paper mainly research how to acquire the images' depth information by binocular stereo vision, under the application background of recognizing the targets from the crowd scenes. In this article, I will specially analysis the key technology of binocular vision, including the camera calibration, stereo matching and 3D reconstruction. Then, a hardware platform is designed based on DSP.First, for the warts and pimples of the existing calibration methods, this paper propose a method based on wavelet neural network and come up with the calibration model, while using particle swarm optimization to improve learning algorithm. And compare the results of this calibration with BP neural network method. Experimental results show that:Camera calibration of binocular vision system based on Wavelet neural network achieves a higher real-time, calibration accuracy and generalization capabilities.Second, mainly research the matching algorithm based on Dynamic Programming and the algorithm based on Graph Cuts. And for their advantages and defects, a new stereo matching methods based on Image Segmentation is put forward. At the same time, the matching model, theoretical analysis and specific operations are also given. It is proved by experiments that this algorithm is efficient and reliable, and can improve the matching speed, solve the problems of low matching-rate and horizontal streaks of disparity map.Then, I will compute the space coordinates of the points in the world coordinate system which correspond to the images, according to the results of the stereo matching. And then,drawing the 3D scenes model by using the linear interpolation method and OpenGL which is a 3D graphic API with powerful drawing and display. It has got good effects of 3D reconstruction.At last, the paper introduces the design of hardware platform for the binocular stereo vision system, and presents the hardware circuit diagram. Dialing, testing and driver programming are also finished. Then, the software of the binocular stereo vision algorithms are transplanted to this platform and are tested. Finally, the goal of obtaining scenes'depth information is realized.