Automatic Target Recognition And Detection In Complex Environments

Automatic target recognition and detection is a hot and difficult of the researchersin a complex scene, it has a very good application prospects in the military and civilian,many countries have invested a lot of manpower and resources, expectations a place inthe increasingly complex military environment and public safety, and maintains theiradvanced nature.This thesis had an in-depth study of the automatic target recognition methods incomplex scenes, to summarize the work of predecessors, and proposed a set ofautomatic target identification adapt to different scenarios, the concrete work andachievements are as follows:(1) Researched the basic fractal theory, and proposed the simplified fractaldetection algorithm and the fractal algorithm based on the target gradient, it couldgreatly improve the efficiency of the algorithm, and achieved good fractal detectioneffect for the faint edge of the target, because the fractal algorithm segmentation targetcontaining empty area, this thesis proposed a filling algorithm based on target the outercontour, get better target segmentation.(2) Researched canny edge detection algorithm and generalized edge detectionalgorithm, and linked the detected edge, get a more complete object contour edge,because the complexity of the linked edge image, the target edges and backgroundedges difficult to distinguish, this paper through anti-color the edge and restricted thetarget area, got the ideal segmentation of targets, it’s very close to the actual target.(3) Researched the gray area feature and binary edge shape feature of the target,extracted the Hu moments and energy feature of the target gray area, extract Fourierdescriptors of the target binary edge, and calculated the feature matching error in thesequence of frames, got the error threshold of every feature, laid the foundation forcontinuous tracking identification of the subsequent targets.(4) Researched the Hough forest target detection algorithm in the machine learning,and proposed the cascading Hough forest projection multiple target detection algorithmbased on the single-class object detection, achieved good results through testing the actual data, and validate the detection performance of the algorithm.