Image Marker Recognition Based On FPGA

With the wide application of high frame rate and high resolution digital cameras,higher requirements are placed on the accuracy of videogrammetry measurement and the real-time processing efficiency of images.The research on real-time recognition technology of image marker point coordinates is of great significance to the engineering field that needs to quickly obtain high frame rate and high resolution video measurement data.The existing real-time recognition method of marker point coordinates is divided into multi-core parallel and image segmentation parallel,which are serial recognition mark points,but it does not really achieve parallelism in algorithm.Therefore,this paper focuses on FPGA-based image marker point real-time recognition technology research.The image parallel segmentation method of video measurement based on FPGA is studied.The method of coordinate image prediction based on time series image is proposed.The parallel segmentation architecture of video measurement image based on FPGA is established.That is,the marker point estimation coordinates are obtained by the marker point motion estimation model,and the FPGA realizes the parallel segmentation video measurement image according to each estimated coordinate,thereby achieving the purpose of extracting the marker points in parallel.The method not only shortens the time overhead of searching for large-scale mark points,but also realizes parallel image segmentation and parallel recognition of marked points,so the real-time performance is good.The eight-step tracking method and the gray center of gravity method FPGA implementation technology in the process of marker point recognition are studied.The efficiency of the FPGA algorithm of the eight-step tracking and gray center of gravity method is crucial for ensuring real-time performance,which is the difficulty and focus of this research.An eight-neighbor search method for incremental update is proposed.The eight-neighbor search is divided into two parts: the first part acquires the connection point in the searched eight neighborhoods;the second part searches for the connection point in the newly added eight-neighbor domain.Then the two parts of the connection point are put together to form a complete array of connection point,and the coincidence point is in the first place.The method not only avoids repeatedly searching for the eight neighborhoods,but also eliminates the need to distinguish the contour coincidence points,which greatly shortens the eight neighborhood retrieval time overhead and the resource overhead of the contour tracking module.According to the parallelism of FPGA pipeline,when the gray weight value is calculated,the original gray is again subjected to gray stretching and binarization,thereby discriminating the pixel points of the marker point and the pixels points of the non-marking point,thereby saving the stored binarized image data Resource overhead.FPGA resource optimization and time consumption prediction model for video measurement images.Studying the prediction model of FPGA resource optimization and time consumption,under the condition of given FPGA resources and known video measurement detection image resolution,according to the size and number of marked points,considering the optimal FPGA resources and real-time conditions,Determine the number of points processed in parallel to get real-time processing efficiency.The FPGA-based marker point image coordinate extraction platform was built and experimental research was carried out.Design system hardware experiment platform,develop PCIe communication protocol,PC real-time monitoring software.The host computer is responsible for real-time monitoring display,parameter setting,marker point coordinate storage,FPGA is responsible for processing image real-time processing,and the experimental platform real-time processing capability reaches 2 GB/s.