Research On The Interest Target Detection And Tracking Technology Aerial Images

Interest target detection and tracking in aerial images has become the focus in the field of computer vision research in the past ten years. Aerial photograph has a broad prospect in both civilian and military fields such as seismic surveys, emergency relief, nuclear radiation detection, military surveillance and Diaoyu Islands activities. However, there exists relative movement between the the camera and the target in general aerial videos, because the two move at the same time. Therefore, in the pictures obtained, there are irregular background movement which becomes dynamic background. Consequently, it is difficult to effectively detect the actual movement state of interest target in such a complex background. So, the target detection and tracking technology in the traditionally static context can not be applied in the aerial photograph. Besides, aerial videos have such features as the serious random movement of the vehicle, a wide shooting range, a few target features, small target and many false targets. What’s more, the image quality is also vulnerable to the interference of noise and external environment. As a result, it becomes increasingly difficult to track and detect the target. Therefore, the study of interest target detection and tracking technology in aerial images is quite prospective and challenging. The contents of the thesis are as follows:For the feature that aerial vehicle and the target move at the same time, based on the selected feature points, an appropriate matching algorithm in combination with wrong match pairs elimination algorithm is adopted to complete feature matching. As a result, background movement compensation can be quickly and accurately achieved. Then, frame difference is used to detect movement state. In order to remove the interference of noise and false targets, morphological filtering technologies are adopted to further the accurate detection of the interest target. After the study of Mean-Shift and its priori theoretical knowledge, an optimized target model based on traditional Mean-Shift algorithm is adopted and the local effective updates are carried out on the target model for the purpose of enhancing its real-time. Then, after the study of the particle filter and its priori theoretical knowledge, key concepts in the particle filter tracking technology are put forward, namely how to find the location weight and state of each particle in the the center of the target and how to choose an appropriate method to construct a model. In this thesis, particle filter and Mean-shift algorithm are fused and the update template in Mean-Shift is utilized to select the most efficient particles. Then, the model will auto-adapted adjust the number of the particles according to the changes in aerial video images, which can improve the tracking efficiency without losing the immunity and track the target at last.