Research On Moving Target Detection Techniques Based On Visual Attention

Human visual attention mechanism is one of the most important means to Sensing the outside world. Using mathematical model of visual attention mechanism to simulate the human visual perception system is a hot research in the field of computer vision. Researching on computational model of visual attention is not only helpful in understanding the working mechanism of human visual system but also has important application in image analysis and understanding, moving target detection, computer visual, image data matches, Image Compression, and etc.This dissertation proposed an improved visual attention model in the study of human visual perception mechanism and the theory in related field of computer vision. Based on the Itti model, integrating the profile feature maps can improve the recognition ability of the model. A kind of adaptive merge strategy is proposed in this dissertation according to the human retinal cells in different light circumstances on the sensitivity of the color and brightness changes in the characteristics. It is more in line with human visual characteristics. Due to the limited of Itti model of lacking dynamic attention, this dissertation integrated the motion feature to make it applicable to the attention of the video. The experimental results show that the proposed visual attention model improves the ability to identify, extends the range of applications and better simulates the human vision under different circumstances.This dissertation designed a 32-bit vector processor for the features of computing the volume of data, a large number of repeat operations and low data coupling. Using the model of timed Petri Nets to solve the structure dependences, data dependences and control dependences in the pipeline of vector processor. The specific arithmetic of working out the stall periodicity needed by the pipeline is presented. According to the state of usufruct in the modeling, it educes the condition had by data forward. By this arithmetic, calculating the stall periodicity to resolve conflicts. With the analysis of simulated timing graph of vector processor instruction, the scheme is validated. With comparing the time of the same algorithm running on images on different platforms, the vector processor in image processing has obvious advantages.This also dissertation proposed a kind of moving target detection algorithm based on visual attention model. Using the visual attention model preprocesses video frames to find regions of interests as candidate target detection areas and matching the candidate target regions with color and shape matching algorithms. Experiment results on tennis video frames have shown that on the premise that the moving target is simplistic in color and shape and is significant in background is the algorithm proposed by this dissertation can detect the moving target effectively...