Based On Morphology, Weak Target Detection And Tracking Technology

The small dim target detection and tracking technology of video image sequence is the key field in people's research. It is of great value both for military and civil uses.The dissertation introduces the model and parameters of small dim target from three aspects: size, grayscale shape and moving feature, then briefly analyzes some specific difficulties in detection and tracking. Next, some usual experiences and methods accumulated by people are also discussed. In this dissertation, the information in time and spatial are joined to produce system flows.The main idea of this dissertation is realizing target detection and tracking using the methods of digital image processing in mathematic morphology. The goal to this problem is to develop proper background estimation and tracking algorithm. In background estimation, Tophat transform and opening by reconstruction are two popular techniques. In this dissertation, it derivates and introduces the principles, properties and realizations of these two techniques. Then, a comparison between these methods is made in both performance and calculation. Structure element is another vital factor in morphology, since different structure elements will bring huge differences in background estimation. So, the selection skills of structure element is also discussed later.In order to reduce the calculation in tracking stage and the number of false target tracks, it designs a special post processing system before tracking stage. With known shape information of the target, the system selects proper structure element to form two markers by eroding original images. After the reconstruction from these markers, the false targets that is larger or smaller are greatly eliminated. In the next step, ultimate erosion is applied to shrink targets, since it simplifies tracking algorithm.For the binary images after background suppression, the signal to noise ratio of neighbourhood is compared to measure the performance of background suppression methods. A theoretical detection curve can also be compared under different signal to noise ratio. The execution time of different code is selected to measure the calculation.At last, two kinds of image sequence are used to verify the system. For the first one, a series of moving targets is added to constant background under the condition of dim and small. That represents the ideal case after a perfect background alignment. For the second one, it is a real time missile launching image without background alignment, which has some background moving feature. On the basis of the second sequence, the effect of some parameters in tracking stage can be found out.