Research Of Key Technology And Application On The Stereo-image-pairs Capture Based On Embedded System

People always have efforts to make the computer have the capability that cancognitive around environment of information through one or more images. Thiscapability will enable the computer perceive the geometric information of objects inaround environment, including their shape, position, attitude, movement, and candescript, store, identify and understand the these objects. All these are based on imageprocessing, and the image capturing is the foundation of subsequent procedure. Theimage capturing pointed here not only refers to obtain “original image” throughphysical device, but also convert the “original image” to stereo-image-pairs that canextract three-dimensional information easily.There are two methods for gettingstereo-image-pairs:the first one is to capture original image by physical device,andcalibrate device to obtain the device`s parameters and geometry information, thenconvert the original image to stereo-image-pairs. The second one is convert the singleplane image to stereo-image-pairs through software transformation. This paper hasresearched and implemented these two methods.For the first method, this paper has researched the relevant technologies andmethods of camera calibration based on computer vision theory and focused on stereocamera calibrating methods and steps. This paper implemented a based onembed-platform stereo-image-pairs acquisition system. This system can do the left-rightsingle camera calibration and optimization results, apply the calibration result to imageundistortion, It also applies the stereo vision`s principle of epipolar constraint to rectifyleft-right image, and output stereo-image-pairs finally. Embed system is widely used inall aspects of our social life by its small size, fast speed and flexible application, etc.This paper realizes the complex camera calibration process in an embedded system, anduse the nonlinear calibration method to rectify left-right image. Experimental resultshows that the calibration result is stable, high precision and has better real-time. Forthe close-range photogrammetry and three-dimensional reconstruction this systemprovides essential data and laid a good foundation.For the second method, we researched the uniform distribution and normaldistribution generated random-variables-matrix`s images by analyzed the Hou-method.The experiments and analysis shows that compared with others this method is simple,but has limited use scope.