Research On Image Matching Algorithm For Programmable Vision Chips

Image matching technology plays the core role in processing of the image data. It’s the basis of many applications on machine vision such like object recognition、 attitude determination、image registration and moving target detection and tracing. Because of feature point matching has the advantage of strong robustness and high matching efficiency, in recent years, the research of image matching is especially focus on feature point matching. This thesis mainly research on image matching algorithm for programmable vision chips:(1)This thesis has analyzed the SIFT (Scale-Invariant Feature Transform) and BRIEF (Binary Robust Independent Elementary Features) feature matching algorithms. But the original SIFT algorithm has large of complicated floating-point multiplication, which is not suitable for vision chip, and the BRIEF algorithm is inefficiency on the extracted feature points. To analyze the above characteristics of these two algorithms, this thesis provides a new algorithm.(2)This thesis has analyzed the architecture of the programmable vision chip. Therefor provides a new feature matching algorithm based on vision chip. This new algorithm Improving the extracted feature points via SIFT algorithm to suitable vision chip. Using an improved binary descriptor by the BRIEF instead of complicated SIFT descriptor. So the new descriptors can use Hamming distance to measure the similarity.(3)This thesis has designed an experiment to test the new feature matching algorithm and to analyze the result of experiment. The result shows that OBF-SIFT can accomplish the feature matching between two pictures and also has a fast speed. The experiment measure the different parameter setting effects on feature matching result, and provide the best parameter setting of the result.To sum up, this thesis starts with studying key problems of the feature point matching algorithm and vision chip, and then improves an algorithm based on the programmable vision chip. Its real-time has been highly improved and has been confirmed its availability on the vision chip.