Design And Implementation On Green Crop Video Stream Depth Map Using Binocular Vision Technology Based On FPGA

Stereo vision is one of the important research topics in the field of computer vision.The theoretical basis of stereo vision is a bionic to the human eye.By fixing two identical cameras together and keeping them at a certain distance,then shooting the same scene,and finally,using binocular stereo matching algorithm.The disparity value of the image is calculated,and the internal and external parameters obtained by the two camera calibrations and the distance between the cameras are combined to extract the three-dimensional information and the real shape of the actual scene.Compared with other three-dimensional information acquisition methods,stereo vision has the advantages of simple equipment,low power consumption and non-destructive measurement.Therefore,it has important application value in the fields of robot autonomous navigation,medicine,virtual reality and scene reconstruction.However,at present,stereoscopic vision technology is less used in practical systems because of the problem that it is difficult to satisfy both real-time and accuracy,and the generalusing processor will be rather weak in the face of the problem,and there is less video stream-based accelerate based on FPGA.In this thesis,based on the real-time and accuracy requirements in real field applications,and the software algorithm and hardware structure are considered comprehensively,the stereo matching algorithm of BM and the green crop region extraction algorithm are implemented based on FPGA.Experiments on the Middlebury platform show that the BM stereo matching algorithm has a higher advantage in high-texture images.At the same time,according to the characteristics of the algorithm,a structure that can execute the BM algorithm and the green crop region extraction algorithm in parallel is designed in the FPGA,and the two steps of the BM algorithm and the two steps of the green crop region extraction algorithm are all run by the pipeline structure.For images with a resolution of 1920?1080,the overall running speed of the algorithm reaches 49.75 FPS,and the running speed of the BM algorithm reaches 51.59 FPS,which meets the real-time demand of the field depth information extraction.