Image Matching And Moving Object Detection In Video Monitoring

The video monitoring system designed in this thesis is applied between the train locomotive and carriage. This system mainly performs two functions: For one thing, it is applied to the recognition of the different states (hook up or not) of the train hooks based on the image template matching, so as to improve the work efficiency of the motorman when he hooks the train hooks; for another thing, it is applied to the moving object detection between the train locomotive and carriage lest the train hurts people when it starts.On the research of the recognition of the different states of the train hooks, a method including two steps from coarse matching to fine matching is proposed: The normalized product correlation image matching algorithm based on template matching is used for coarse matching, the moment invariants image matching algorithm is used for fine matching, and the templete is updated adaptively during matching. After the mechanical structure of the train hooks analysed, the different states (hook up or not) of the train hooks are classed, and then the two steps coarse-fine matching method is adopted. The normalized product correlation matching algorithm is fit for the recognition of the hooksed state of the train hooks, however, it fails when the image has been translated, scaled and rotated, which are fit the fine matching step using moment invariants as the matching feature. Experiments show that combined matching algorithm is very stable with high accurate rate.On the research of moving object detection between the train locomotive and carriage, a moving object detection algorithm combined frame difference detection algorithm with background difference detection algorithm is proposed. For the first the paper gives the opposite processing methods for the image noise brought by camera shakes, and then the strongpoints of the two detection algorithms including the stability of frame difference detection algorithm and the accuracy of background difference detection algorithm are combined, so as to detect and extract the object region completetly and accurately.The simulation experimental results demonstrate the practicability of the moving object detection and the advantage of the proposed algorithm.