Research On Real-time Positioning Algorithmbased On FPGA

Positioning based on vision sensor is a highly developing positioning method in recent years. The principle is to obtain an image of the scene around through the camera and use some natural or man-made features of the scene and the image processing method to achieve the surrounding environment model, then to determine the relationship between target object and the camera itself. The method has important use value in the areas of real-time digital photogrammetry、robot navigation、active vision positioning of aircraft, and so on. It is also the more popular research topic currently. In the area of vision sensor positioning, monocular orientation avoiding the defects of the small visual field of stereoscopic method and difficult stereo matching, is gradually becoming an important branch of the spatial location area with the advantages of simple structure and less calibration steps, etc. However, while monocular positioning has made remarkable progress after years of research and exploration, but there are still many obstacles which can’t be overcame still now. The current biggest difficulty lies in extracting of feature point coordinates, the main problems are:(1)The algorithm always changes with the variation of the environment around, which may do a greater impact on the correct extracting of the feature points;(2) Currently, the most algorithm of the extracting of the feature points are related to converged applications of multi-image processing technology, and the use of a unified theory system to accomplish all aspects of the algorithm is rare;(3) The majority algorithm of the feature extraction can only be handled by computer which is not portable,and we also can’t achieve real-time result because of the complexity of the algorithm. The thesis will use mathematical morphology and segmentation as the theoretical guidance, to focus on the research of portable and real-time implementation algorithm.Firstly, this thesis reviewed and summarized the current individual visual localization algorithm and proposed the architecture of real-time positioning algorithm of monocular camera and presented insufficient aspects of current coordinate extraction methods with a brief analysis from target recognition, object extraction, to 3D reconstruction. This thesis was committed to the development of a new localization algorithm and to achieve accurate real-time implementation positioning aspects by way of FPGA.Secondly, this thesis dialyzed the relationships between the various coordinate systems which lay a solid theoretical foundation for the development of the algorithm of the third chapter through stereo vision measuring principle of the second chapter.Thirdly, this thesis illustrated the coordinate extraction defects of current various methods, such as edge wavelet transform modulus maxima detection, Harris corner detection algorithm, Harris-Laplace corner detection algorithm, etc. By analysis of the characteristics of the image data, the thesis proposed a new coordinate extraction method based on morphological theory and image segmentation technique. By calling various software such as MATLAB、System Generator and other tools for the development of algorithm.Finally, in order to achieve real-time positioning effect, the thesis selected Xilinx FPGA chip as a hardware platform, and lastly, demonstrated the results of software and hardware platform.