Study On The Algorithms For Small Moving Object Detection In Complex Background

Small moving object detection from complex background is a very usefulwork, which can be used in martial navigation, bomb tract recognition,trafficpeccancy detection and other fields. But in some fields such as Infrared guidingmissile, we need to capture the targets as fast as possible, but the small targetusually is drowned in all kinds of complex background and noises only for thedistance between infrared sensor and object is far. The point targets have no solidgeometrical feature and the dimension is comparatively small against the largebackground, the SNR is so low that it's difficult to detect the point targets only insingle frame. So, some detection algorithm based on edge or energy accumulationcan't acquire. We came out a series of small object detection algorithm accordinga large amount of Chinese and Foreign papers and the correlation of small objectsin adjacent frames. The small moving object has the specific feature against the complexbackground: 1. Small dimension. 2. Weak contrast. 3. High interference fromatmospherics. 4. Strong correlation of the moving target in adjacent frames. We can remove large area background and keep the point targets by adoptingTophat Morphological Filter, Median Difference Filter, Wavelet Difference Filteror High Filter based on reconstruct of high frequency Wavelet coefficients andother high Filters. So, the main aim is to design a high pass filter. The Filtered frame is a frame that only include large area zero backgroundand high frequency point noise including the point target. Then, we canaccumulate the deference frame. The point target represent as point tract becauseof it's moving, while some point noise can counteract for it's random, other pointnoise represent as high randomly interferential points. Our experiments can prove it's valid in small object detection. In the sum frame, we can adopt bi-directional chain and recursion algorithmto detect according to the high correlation of the moving target. In every node, theposition and scale information are stored, centered as the current point, search the - i -吉林大学硕士学位论文 next point in the semi circle which radius is measured by the multiply of flaptime and the project velocity in imaging plane. Making the number of tract pointas threshold, when the length of chain amount to the threshold, output the all thechain points, which is the small moving target's tract. The following figureillustrates the base principle Image Multi-Frame High filter: Tract- sequence Tophat judge Median difference Wavelet filter Sum Figure 1.1 .1 The fundamental principle of Tophat morphologic high pass filter Morphology is a new image processing method, which make erosion,dilatation as it's base. Erosion has the effect to shrink the original image, whiledilatation have the effect to expand the original image. The open and closemethod can be derived from the erosion and dilation method, which each can bedefined as to erode first then dilate and to dilate then erode. Open method hasthe effect to polish the image outside border, while close method has the effectto polish the image inside border. Then Balckhat and Whitehat method can bedefined based on the open and close method, which are defined as Tophattransform. Whitehat Transform: The difference image of the original image and themorphologic open image, which is defined as: Whitehat(f) = f - f°g Blackhat Transform: The difference image of the morphologic close imageand the original image, which is defined as: Blackhat( f ) = f ? g ? f...