Research On Key Algorithms Of Moving Target Detection And Tracking In Aerial Video

The moving target detection and tracking in aerial video is the key technology of UAV reconnaissance,monitoring control from an aerial base and detecting geographical disasters.It has great significance to the development of national defense and national economic construction.But it is a challenge work to detect and track targets in aerial video as the characteristics of aerial video include smaller target sizes,larger scale variations,more frequent occlusions,lower resolution and more complex scene compared with common videos.In this thesis,the key algorithms of moving object detection and tracking in aerial video are studied deeply aiming at improving the accuracy and real-time performance of the algorithm.The main work of this thesis is as follows:Aiming to the characteristics of moving background and complex scene in aerial video,a kind of moving object detection method based on background compensation and Temporal-Spatial information are proposed.Firstly,the ORB features with good performance are selected as the feature points in the background compensation of aerial video,then the spatial constraint distribution is used to make the extracted ORB points distribute in the image frame evenly.Furthermore,the RANSAC algorithm is used to estimate the homography transformation model of the background for motion compensation.In the moving target detection,firstly we using frame difference method to detect the moving targets.As the detection results of the frame difference method are generally exists "empty" and noise,in order to get more accurate extraction results of the moving target,the spatial information are combined for further target segmentation.The experimental results on the DARPA VIVID dataset show that the algorithm proposed in this thesis can detect the multiple motion targets accurately in aerial video.The strategy of the Temporal-Spatial information enhances the robustness of the detection algorithm.Aiming at the scale variations of moving target in aerial video,a kind of tracking method of adaptive scale change based on KCF algorithm are proposed.The possible scale changes of the target are detected in each frame and the value meeting the threshold is regarded as the current optimal matching scale.This scale detection strategy with an adaptive threshold can reduce redundant detection and improve the real-time performance of the tracking algorithm under the premise of ensuring the tracking accuracy.Meanwhile,a long-term tracking strategy based on KCF algorithm is proposed for the occlusion or target out-of-view and re-entering the field of view in the aerial video,which introduce the mechanism of abnormal motion judgment and the re-detection module,and to re-detect the target based on multi-feature fusion matching.The experimental results on the DARPA VIVID aerial dataset demonstrate that the long-term tracking strategy proposed in this thesis improves the robustness of the tracking algorithm.Our method achieves an impressive performance compared with several state-of-the-art trackers such as TLD,CSK,KCF and SAMF in aerial video tracking.Compared with KCF,our proposed tracker achieves a 24% improvement in precision and a 35% improvement in success metrics.Furthermore,our approach is capable of tracking aerial video in real-time while running at more than 70 frames per second.