Research On The Image Real-Time Acquisition, Storage And Image Processing System

The real-time performance of the image process is consisted of real-time image data acquisition-storage and real-time image process. The real-time image data acquisition-storage and real-time image process are conflicted within the usage of the compute memory. The performance of the real-time task is deteriorative especially when the work is hard. So in most conditions the real-time performance can not be satisfied.In order to solve the confliction, the real-time image data acquisition, storage and image processing system is designed which can satisfy the real-time performance. In this dissertation, the structure of the system and the design of the hardware are described and the details are discussed. The soft-ware of the system is designed using the VC++ development and the SDK function library supplied by the IO industries.During the image processing, the features of object are extracted primarily. The object features can demonstrate the characters and status of the objects in the image. Also the object features can effectively reduce data to be processed. So feature extraction is a validate technique to implement the real-time image processing. The features extracted are mainly edges, lines, corners and moment invariants. The edge is presenting the contour of the object; the line is a special kind of edge, and can provide the image information for the camera calibration and vision tracing; corner and moment invariants are the stable geometrical features and both of them have the invariability when translation, moving and rotating. The corner feature is widely used for the three-dimensional matching, target tracing, and position estimation. The moment feature is widely used for the target recognition. In the dissertation the methods for the feature detecting are researched, compared and analyzed through experiments.The methods for the geometrical target recognition are researched based on the extracted features. First, the targets are recognized by the amount of the corners. So the corner detectors must be stable. Then the least square method is applied to detect corners. Second, the moment invariants extraction experiments indicate that the moment invariants have clustering. Then the fuzzy clustering recognition theory is applied to recognize the target. Finally, the real-time processing experiment is designed, and the real-time processing speed is measured.