Research On Stable Tracking Method For Air-to-ground Moving Target Under Occlusion Conditions

At present,air-to-ground surveillance has gradually replaced ground-to-ground surveillance of fixed position due to its large field of vision,strong maneuverability.In addition,the surveillance system based on UAV(Unmanned Aerial Vehicle)with the characteristics of low cost,operability and safety has been widely used in the civil applications,such as traffic monitoring and disaster search etc.Object tracking is the key technology for air-to-ground surveillance to achieve moving target reconnaissance and tracking strike.It is defined as by locating the target in each frame,its position information can be given to the observer for the further understanding of its behavior.However,because of the background surroundings,the target will be occluded heavily and disappears in the field of vision that affects the tracking results.Therefore,the requirement of stable video tracking of ground moving targets with occlusion handling arises.This paper focuses on the anti-occlusion target tracking algorithm.Its main contribution are summarized as follows:(1)An improved Camshift target tracking algorithm is proposed for simple background.Since Camshift algorithm is used to model the appearance of the target,background interference should be minimized and foreground information should be fully utilized.Using OTSU threshold segmentation algorithm,the target and background in simple background can be separated.Therefore,after rough selection of tracking region,the foreground region can be refined and the minimum region that contains the whole target can be obtained.Then this region is used as the initial state to start Camshift.Experiments on two sets of aerial video sequences under simple background show that the improved algorithm effectively reduces background interference and greatly improves tracking accuracy.(2)An improved long-term correlation tracking(LCT)algorithm(ILCT)is proposed.LCT can effectively handle scale variation,illumination variation and background complexity etc.,while tracking failure occurs when the target is seriously occluded.Therefore,an occlusion trigger and an object re-detector through entire image are added on the original LCT tracking framework.This occlusion trigger is designed on the characteristics of the response curve of correlation filtering under occlusion.If the response curve satisfies the occlusion trigger,the tracker and the update of filters are ordered to stop and the re-detector is initialized.The result that meets the re-detection confidence will let the tracker resume as a new initial state.Otherwise,it will enter the next frame to continue re-detecting.Experiments are carried out in several video sequences with severe occlusion.ILCT is compared with other state-of-the-art trackers and the qualitative and quantitative results show the robustness and accuracy of ILCT.(3)An airborne tracking experiments and simulation system is designed and established,which consists of hardware platform and software platform.The hardware platform is based on DJI Spark drone.It is used to capture aerial videos with severe occlusion as data source for the software platform.The software platform is based on MATLAB 2016 b.ILCT is used in the platform to track the target in aerial video that the effectiveness of ILCT is verified.