Research On Moving Object Detection Based On Visual Attention And Improved Bacteria Filter Algorithm

The visual attention mechanism is a model of human visual system which is used to catch and process information, it has efficient data filtering capability. So that, human beings can pay more attention to a small number of remarkably objects from complex scenes quickly. It is very important to introduce visual attention mechanism into the field of image processing which needs massive data calculated. This mechanism has theoretical value and practical significance.The purpose of moving target detection is split out from the video will be of human interest sports area, computer vision research and critical step. In recent years, moving target detection widely used in intelligent monitoring systems, medical image analysis and video image compression and transmission, and other fields. Rapid, accurate detection of interest moving target is a moving target classification, tracking and behavior analysis of the basis of the subsequent processing.This paper first introduced the research of visual attention algorithms, and proposed an improved attention model. The model on the basis of the Itti model, adding motion information characteristics, making it suitable for dynamic image. We proposed an adaptive feature merger strategy based on the character of the human visual system that pays more attention to the movement area. This article also reviewed three types moving target detection algorithm:optical flow method, the inter-frame difference method and background subtraction method, analyzed the advantages and disadvantages of each method. Then, we proposed a new moving object detection method combined visual attention model and improved particle filter. Firstly, we preprocess video sequence to detect the interested areas of the image by using visual attention model. Then, the moving objects are obtained in the interested region by using improved particle filtering algorithm. This paper used block-by-block as processing unit improved running efficiency, noise immunity and resistance to illumination change capacity of the algorithm. In addition, this article utilized the cosine distance, can be able to make heavy use of features, to calculate the similarity between the particles and the current image block. Experimental results showed that this method substantial increased the speed of system operation, be able to establish a good background model and detect moving objects accurately.