Research On Detection Technology Of Moving Target Based On Saliency

Peoplecan quickly find the interesting region even in the complex scenes. This is the so-called vision selective systemand the region is called salient areawhich is general sparse, strong contrast. Salient object detection gets the foreground different with the surroundings by imitating the human vision selective system. And it’s mostly as the image preprocessing module for providing rough location of target. It has been widely used in image compression, human-computer interaction,object tracking and recognition and so on, and its research and application valueis undeniable.Traditional moving object detection methods have more or less defects, such as background subtraction method for slow moving target and frame difference method for contour detection is not satisfactory, optical flow method is easily affected by noise, illumination changes and so on. The above method has poor performance in the complex or changing background, but salient object detection can solve the problemsbetter. It can provide rough position through the spatial saliency map of the current frameand then through the temporal saliency map of the adjacent frame further refined, therefore the effect of noise, illumination, background is small. Existing saliency methods are mostly based on static images butnot the video frame. In this article, we combine spatial and temporal saliency map which can effectively deal with the video frame. The spatial saliency map obtainsby improving thefrequency based method FT which is improved by pretreatment, two color space and center-surround contrast operator to optimize the saliency map. The background suppression ability is better and also conducive to highlight foreground. The temporal saliency map obtains by optical flow method to extract motion vectors of neighboring frames, andwe put forward the weighted contrast operator which will pay more attention to the area closer to the centerfor the deficiency of center-surround contrast operator. Then select the relative optimal scaleand thus to gain the temporal saliency map. Finally reasonable fusion method can effectively combine the advantages of spatial and temporal saliency map for a more complete object, andselect the normalized sum fusion as the most suitable method of this paper through the experiment.