Study On Stereo Matching In Micro Stereo Vision System

With the development of science and technology, the need of micro system is more and more urgent in many technical fields, such as various operation of cell and polymerized substances, micro surgery, scanning probe microscope (SPM), butting optical fiber, fine manufacturing etc. With the development of micro-technology, micro mechanism, which has the character of micro size or micro motion is new high technology from microcosmic point of view understanding and reconstructing the world, micro mechanism technology is important means for researching nanotechnology, so the micro stereo vision(MSV)techniques are demanded urgently in microcosmic domain.Stereo microscope and CCD sensors make the main components of a MSV system. The image information collected by CCD sensors is transformed as digital image information by image capture card. The digital information is easy to process by PC. This paper introduces the stereo matching techniques in MSV system, including image pre-processing, feature abstraction, stereo matching and depth information acquiring.In order to obtain images easy to detect objects' edges, images have to processed by the techniques such as Guass smoothing and image enhancement. And then most of the noises in images are eliminated successfully. Several edge detection operators such as Roberts, Sobel, Prewitt, Krisch, LoG and Canny act on same images. Comparing the output images, I select Canny operator as optimizing edge detection operator.This paper emphasizes on stereo matching. Firstly, selecting objects' edge pixels as matching elements. For the sake of sparseness of edge pixels, the matching area may big enough to find an accurate corresponding edge pixel. After this processing, we could obtain most edge pixels' corresponding pixels and maximum of disparity. In order to make up for the sparseness of edge pixels, a new matching algorithm named Census is introduced, which based on neighbor pixels' correlativity. Census's searching area is limited in a square whose side length is two times maximum of disparity. As the searching area's lessening, the operating speed increase greatly.