Research On Moving Weak Small Target Detection In High Clutter Backgrounds

In recent years，the weak moving target detection technique is one of the mostcentral issues，including computer vision, image processing and pattern recognition. It iswidely used in various fields such as precise navigation and astronomy observation.Because of the complex background of the weak moving target detection, it is easy to beoverwhelmed by the noise and background interference. Moreover, the self-movementof camera system leads to the obvious location differences during the two consequentimages. Therefore, it is very important to study this issue, no matter for the theory or thereality value. in order to solve these problems, much work has been done and it can besummarized as the following:1. A reliable feature extraction algorithm is proposed. The algorithm firstly smooththe image by Gaussian function and then select the feature points which are reliable inall the smoothed images. It is shown that higher position accuracy and betterperformance against to noise can be achieved when compared with Moravec and Harris.2. An improved ICP algorithm has been proposed. In order to solve thedisadvantages of poor robustness and real time performance of original ICP, we use thesoft shape-context to compute the feature similarity and assign different weights to pointpairs. Besides, the Adaptive Dual K-Dtree strategy is used to accelerate the speed. It hasbeen shown that the improved ICP algorithm is more robust and fast.3. A compensation method of moving background based improved ICP algorithmis proposed. In order to compensate the position differences caused by theself-movement of camera system, we use the improved ICP algorithm to register thetwo consequent images. So we can align the background and detect the real movingtarget more easily.4. A systematic TBD algorithm for weak moving target detection is realized byusing the techniques described above. Based on the moving characteristics of smalltarget, we use the multiple images to form the motion orbit and apply the EKF for targettracking. By doing so, we have effectively solved the problem of weak moving targetdetection in complex background, occlusion and noise environment.