Research On Identification And Position For Binocular Vision Based On Embedded System

The thesis studies binocular stereo vision technology based on embedded system, we designed an embedded binocular vision based on ARM processors. The main function of the camera is to solve the conventional binocular camera with personal computer as processor brings high power, shortcomings expensive and difficult modular applications. With the development of embedded technology, the current ARM processor computing speed has been able to meet the processing speed of the binocular vision system requirements, and low-power embedded chips, small size can also promote the embedded binocular vision application and become the focus of the current research in the robotics and unmanned aerial vehicles(UAV) field.In the course of study, embedded processor is chosen according to project requirement and designs the system hardware. Analysis of the binocular vision image processing algorithms and research several kinds of algorithm characteristic and local matching algorithm is chosen according to the experimental and the specification. Based on the embedded characteristic, designs and builds the embedded software system including the operating system and with OpenCV and QT as the representative of the software library. Finally, the recognition and positioning and real-time performance is verified about the embedded binocular vision system.The research results of the thesis include Samsung Exyson 4412 processor as the core processor, designing serial circuit, ethernet interface circuit, USB interface circuit and other peripheral circuits beside core board and build a complete embedded hardware platform. Research the binocular vision image processing algorithms, through establishing an ideal model of binocular vision, analysis and derivation principle binocular imaging and coordinate transformation are used. And using MATLAB and OpenCV camera calibration compare the results of the calibration. Based on the stereo matching algorithm, and analyzes the different types of matching algorithm principles, advantages and disadvantages, and providing a theoretical basis for choosing stereo matching algorithm later. Construction of embedded vision systems, by comparing the choice of embedded linux operating system to build the embedded development environment, including system boot procedures, the transplantation of linux kernel and constructing root file system, designed to the software application framework platform consisted of V4L2 video capture driver interface, OpenCV vision libraries and QT software libraries, and achieve embedded binocular camera capturing and processing. According to the characteristics of the stereo matching algorithm, finished the binocular vision in the object recognition and position, and verify the feasibility of the system.