Vision Target Tracking Technology Of UAV

At this stage,the development of computer vision technology and UAV platforms has become diversified,and people are increasingly demanding the visual tracking function under UAV platforms.In this paper,most of the tracking algorithms are difficult to meet the real-time,accuracy and robustness requirements in the application of UAV platform.Taking the application of visual target tracking technology under the UAV platform as the research center,the modular form will be used The visual tracking technology is divided into three parts: target detection,visual single target tracking,and visual multi-target tracking.Among them,specific research is conducted on the application of single-target visual tracking in the three scenarios of UAV to sea,UAV to ground,and UAV to air.Finally,the functional results of each part are displayed in the form of visual interactive software.The main research work and results are as follows: 1.Target Detection.The two target detection algorithms based on YOLO v3 and background modeling are studied.By analyzing their advantages and disadvantages,a target detection algorithm model suitable for the application of the UAV platform is established.Finally,the YOLO v3 algorithm is selected as the initialization in the target tracking task.means.2.Visual single target tracking.A variety of related filtering tracking algorithms and classic tracking algorithms are researched,and the visual single target tracking algorithm suitable for different scenarios is selected by analyzing its advantages and disadvantages.In order to adapt to the complex tracking tasks in the air-to-air scene of the drone,an optimization algorithm combining YOLO v3 algorithm and FDSST visual tracking algorithm is proposed.The YOLO v3 algorithm is used as an initialization and target tracking error correction method to ensure the robustness of the tracking task Experiments show that the algorithm can well track single target tracking tasks of UAVs in air scenarios.3.Visual multi-target tracking.A multi-target tracking algorithm based on the YOLO v3 algorithm and multiple single-target trackers is proposed;each target tracker is independent of each other,and the corresponding motion trajectory is matched by the Hungarian algorithm.Quantity and position correction methods to ensure the stability of target tracking tasks.Experiments show that the algorithm can complete multi-target tracking tasks with good robustness and real-time performance.4.Design of UAV Vision Target Tracking Software.The software mainly includes four modules: target detection and recognition,single target tracking,multi-target tracking,and drone detection and tracking;different algorithms are integrated into the software in the form of a unified interface,and visual display functions of each module are provided.