Stereo Matching Algorithm Based On Binocular Vision And Its Implementation With FPGA

Binocular stereo vision is one of the hot issues in the field of computer vision technology research. It mainly uses two cameras to shoot the same scene.According to the geometrical principle of different locations of two-dimensional image information of 3D reconstruction, the three-dimensional information of the original scene can be restored. It is an important means to obtain the depth information of the 3D scene due to the way to directly simulate the human eye. Stereo matching is an important research direction in binocular stereo vision and mainly used in the field of robot navigation, hree-dimensional measurement and virtual reality.The improved matching algorithm improves the matching accuracy. Firstly,in the Mini-Census transform, the neighborhood pixels need to be compared with the average values of the pixels in the center and the neighboring pixels, and a series of two digits are obtained.Then the size of the matching window is dynamically adjusted by using the gradient information of the image. Finally, the left and right consistency checking, voting algorithm and histogram are combined to refine the disparity map to get more accurate and dense disparity map.The binocular vision stereo matching algorithms with computational complexity and large amount of data are unable to meet the real-time requirement,so they are restricted in the practical applications.This paper chooses Field Programmable Gate Arrays(FPGA) as the platform of hardware and the Verilog HDL is used to design the functional modules of the matching algorithm such as the image data cache, Census transform, gradient calculation, the left and right consistency checking.This paper chooses Altera's EP2C35F672 of Cyclone II series as the processing core, Quartus II 11.1 as the development tool and Signal Tap II and Modelsim SE6.5 as the debugging tools.The stereo images from the Middlebury datasets are chosen as the experimental subjects and this algorithm is verified the feasibility and effectiveness in the matching accuracy and processing speed.