Design And Implementation Of Measurement Based On Stereo Vision

As digital image processing technology develop very fast, machine vision plays a very important role in many areas of the reality life. Binocular Stereo Vision is one of the important forms of machine vision. It depends on vision difference, use cameras take two pictures of the same object from different places. By evaluating the place difference between two matching key points, it can get the 3D information of the object.Binocular Stereo Vision can perform on the plant of cameras with PC, or it can perform on VLSI. Because of PC is inconvenient to take,and VLSI has disandvantages like long developing cycle, expensive cost and complex to modificate. As FPGA develop fast, its processing capacity get much powerful. With the appearing of sopc,it is possible to perform a complex algorithm on FPGA. So it is significant to do binocular Stereo Vision on FPGA. This paper has mainly done following jobs:1. It introduce four kinds of coordinate systems of computer vision. It analysis two kind modules oh binocular Stereo Vision:parallel module and converge module. It is easy for parallel module to evaluate the distance, but very difficult to place the two cameras, so converge module is used in this paper. By infer formulas and analyzing the geometry relationships,this paper gives the formula to evalutate distance. Also this paper analyze the paramers' impact on system's accuracy.2. This paper discuss modules with and without distortion, gives the formula to evaluate distance caring about the distortion. It sketch out popular calibration methods and introduce the Calibration Method of Zhang Zhengyou. It complete the calibration of the cameras by Zhang Zhengyou Calibration Method and matlab toolbox.3. This paper design hardware system of binocular stereo vision, gives the block diagram. Analyze different initial ways of the sensor and acomplish initializing sensor. Cause two camera need to picture synchronously, it is necessary to synchronize the two cameras. It complete the task by operate GPIO. Use DDR3 to store the video data and use verilog to make a gauss fliter.4. This paper complete the SIFT key point detection on NIOSII and matching the key points. Compare them with the matlab result to analyze the accuracy. Evaluate distance by NIOSII and campare it with the actual distance to analyze its accuracy.