Design Of Embedded Binocular Stereo Vision Software And Hardware Platform

Binocular Stereo Vision is an important form of machine vision. It has extensive application prospect because it could work out the relative in-depth information in camera vision simply through image information of the right and left cameras. Although, over decades of years of development stereo vision has been widely applied in robot vision, aerial mapping, reverse engineering, military application, medical imaging and industrial detection through several decades of development, binocular stereo vision has hardly been applied in embedded application due to its technical complexity. Because of its huge calculated amount, its realization needs the on PC platform or collaboration of FPGA on DSP. This embedded binocular stereo vision software and hardware platform is a bold attempt of the technique in embedded application.This paper focuses on the systematic research to embedded binocular stereo vision software and hardware platform as well as the in-depth study to the technical principle and matching algorithm principle of binocular stereo vision. The embedded binocular stereo vision platform adopts TI OMAPL138 DSP(C674X) and ARM(ARM9) dual-core embedded processor(host processor) and two OV7670 image sensors(right and left cameras). The platform totally adopts three AL422 B FIFO chips to realize the cache of the right and left images and matched depth of field. The software part is mainly used to realize right and left image extraction, stereo matching, distance measurement, depth output, etc.The realization process of the embedded binocular stereo vision platform in this paper could be summarized as: 1. principle and physical design of hardware platform; 2. the calibration of right and left cameras by shooting standard image and the acquisition of interior and exterior parameter matrix of cameras; 3. blending interior and exterior parameters into ranging algorithm, transplanting onto OMAPL138 hardware platform, and optimization; 4. the realization of target matching of right and left images, distance measurement of a point in image, and the output of depth information of image.