Research On Moving Object Detection Based On Fusion Of Space - Time

Moving target detection technology is widely used in modern society. As an important branch of computer vision, it has huge development prospect in many fields, including medical field, national defense and traffic monitoring, which is a computer vision preprocessing step by extracting the target area to carry out the target tracking, segmentation and identification of the subsequent operation. In the actual scene, the moving target detection technology is still affected by many uncertain factors, so this is still one of the hot and difficult fields in Computer Vision. While the visual attention mechanism as a burgeoning saliency detection method in recent years, it can effectively detect the interest region by using the human visual mechanism with better detection results, thus inspired many researchers’ study enthusiasm. Although some scholars have already studied the research method which fused the visual saliency to the moving project detection technology, however, the realization of algorithm still exist a lot of difficulties because one of them is the saliency in airspace and the other is the saliency in time domain. In this paper, the main work as follows:In this paper mainly do the following work:(1) The moving target detection technology has been studied, and the principle of the commonly used moving target detection technology including frame difference method, background subtraction method and optical flow method have been analyzed. We also summarized their advantages and disadvantages as well as the applicable scenes. Background subtraction of Gaussian mixture model are analyzed in detail, and the illumination factor is introduced into the Gaussian mixture model, in order to adaptive illumination mutation in real-life scenes.(2) On the basis of analysis, we analyzed the followed principles and common modeling methods, then experiment to compare and analysis the effects of these methods. Among them, in-depth studied the context-aware saliency detection, and apply this model to the Gaussian mixture model, which realize the saliency detection in airspace.(3) This paper presents a spatial-temporal saliency method for moving target detection. Based on the Gaussian mixture model, the input image is divided into a saliency region and a non-saliency region through the saliency detection in airspace, and the modeling process of Gaussian mixture model is influenced dynamically through regional division. The method reduces the waste of resources of the Gaussian mixture model and detects moving targets accurately.The conclusion shows that the proposed method in this paper can accurately detect moving target in high dynamic scenes, and has a strong adaptability to the illumination mutation.