Binocular Stereo Vision Technology And Its Hardware Implementation Research

Binocular stereo vision is an important research subject of computer vision,which can extract the depth information of the three-dimension scene and has been widely used in the fields of army and civilian utilization.Stereo matching is one of the most important problem in binocular stereo vision.In this paper,we have deeply researched how to improve the matching accuracy in depth discontinuities and low textures.We also built the FPGA hardware system to realize the local stereo matching algorithm and measure the distance of the infrared salient object real time.A new stereo based on non-parametric local transforms and effective cost aggregation is proposed and studied to improve the matching accuracy in depth discontinuities and be more robust to the illumination variation.Firstly,a color-based image segmentation algorithm is used to segment the reference image into different blocks.Secondly,Census transform is conducted in both reference image and target image so we can obtain the bit strings.At the.same time,we rectify the disparity range of the pixel to be matched by the disparity of neighborhood pixels dynamically.Lastly,a new matching cost function is used to calculate the cost.The experiments show that our algorithm has better matching accuracy in depth discontinuities and is more robust to illumination variation.In the study of stereo matching algorithms based on low textures,the lack of characteristic difference is the key problem,so we propose and study a new stereo based on the color space and pixel location in the aggregation window.Firstly the low textures of reference image and the target image based on the image normalization and gradient magnitude are extracted,then the cost functions of the corresponding points based on the color space are calculated,last every pixel is given a weight according to its location in the aggregation window.The experiments show that our algorithm can get better matching precision about the colorful low texture stereo image pairs.There is large amount of calculation of stereo matching algorithms which make it difficult to realize on personal computer real time,so we built a binocular stereo vision system with FPGA as the core.According to the parallel calculation and pipeline technology in FPGA,we propose a new method based on Box filter to realize the Census transform.Besides,we implement image sharpening and threshold segmentation to extract the salient object in infrared image and measure the distance.Finally,we realize stereoscopic display using a pair of stereo glasses.The experiments show that our system is accurate and stable.