The Research Of Target Detection Algorithm Based On Visual Attention Mechansim

Target detection is a technology of separating target from background and locating target, taking much time and effort to deal with big data, and lower efficiency compared with man-made method. With highly developed computer technology, artificial intelligence, machine learning, people would consider the target detection as the integrated management of intelligent and automation. The result of target detection is the basis of advanced processed such as target recognition and tracking. It brings affection for following procession. The traditional method of target detection deals with the whole image, but target only occupies the small part of image. The process to search the whole image is complex. Visual attention mechanism makes sit available for finding meaning message quickly from complex scene and abandoning useless message, avoiding traditional method of searching the whole image. The research of application for visual attention mechanism in target detection, computer system for target detection imitate the system of visual attention mechanism. It is a meaningful research to put the visual attention mechanism into target detection.This article introduces the theory、model、attention mechanism and the relationship of attention mechanism with target detection. Details are as follows:(1)Our article studies the development direction of visual attention mechanism and down-top visual attention model. We propose visual model based feature based on model, extracting image color 、 orientation 、 texture feature forming feature saliency map, evaluating our model with tradition model from image focus and the evaluation WTA index.(2)Studying visual model combines top-down model with bottom-up model, and combine visual attention model based on SVM, saliency map from down-top model as training feature for SVM, training the SVM from learning data, then using for test image, this model improve detect precision, saliency map show the position of target.(3)We propose target detection based sift matching from the target based on saliency map, target detection based saliency save time. In addition, we can label the target praised using sift matching.