Small Moving Target Detection Based On Image Sequences

Small target detection in the image of digital image processing has been an importantsubject, but also the focus of many domestic and foreign scholars. Because the lack of size,shape and texture information of small target, it can not be detected in the single image, sorequiring the movement characteristics of the sequence detected. For the shortcomings of lowefficiency of existing algorithms for single frame detection, multi-frame detection algorithmcomplexity, and higher computational ,the paper combined single frame detection andmulti-frame detection algorithm, so as to improve small target detection rate and reduce thedetection time .First, we summarized the regularity characteristics of the three elements of the smalltarget image, target, background and some of the inherent noise, as the basis for follow-upsmall target detection. Secondly, according to results of small target image features, select theappropriate pre-processing algorithm, the paper integrated the high-pass filtering and medianfiltering method to filter the single frame image, then have contrast enhancement. Then, thesmall target images pretreated were segmented with combine algorithm of edge detection andsegmentation image segmentation and the filtered image contrast enhancement, therebyenhancing the target information, weakening the image background information, to improvethe image quality. Then, according to the gradient boundary contact information and provedthe gradient segmentation is more reasonable, we combined edge detection andthreshold-based segmentation algorithm, and simulation results show that this segmentationmethod is almost able to target detection within a single frame image, and the method issimple ,effective and anti-interference. Finally, according to the image small target regularityof the trajectory in the sequence and noise-free trajectory regularity, to detect the small objects with sequence of single frame image . In this paper, superimposed the single frame segmentedimage and selection the targets with eight-neighborhood search-method. Then we could getthe small targets their trajectory.