Development Of Airborne Target Tracker Based On Embedded GPU

With the wide application of UAV detection technology in military and civilian fields,the demand for automatic tracking of ground targets by airborne equipment is becoming more and more urgent.Ground people can control the airborne tracker to finish the monitoring or fixed-point firing tasks automatically.Therefore,the development of airborne automatic tracking equipment has important application significance.Due to the imaging height,ground targets are usually small and difficult to extract effective information for tracking,which adds difficulty to tracking applications in airborne scenarios.The current airborne tracking equipment usually track in the field which has an open view.When in occlusion area,target is often lost,which affects the tracking performance greatly and cannot meet the application requirements.This paper makes research of anti-occlusion robust target tracking algorithm and develops a target automatic tracking device based on embedded GPU.Firstly,this paper analyzes the requirements of airborne tracking system,gives the system performance indicators and designs the framework of the airborne tracking system from the perspective of research and application.In order to improve the problem of tracking failure,through the analysis of characteristics of numbers of tracking methods,the high-speed target tracking technology based on spatio-temporal context learning is chosen as the airborne system tracking scheme from the point view of tracking speed.Then this paper designs the anti-occlusion scheme and studied the technology from the aspects of occlusion time judgment,trajectory prediction and the target reoccupation and makes a reasonable solution.Secondly,this paper uses the embedded GPU platform Jetson TX1 to carry on the development of airborne tracker.Based on CUDA development environment,the application is implemented in Eclipse combined with Open CV in a cross-compiling way to form the high-speed tracking application and the anti-occlusion application.According to the real-time requirements of the tracker,the GPU resources in the embedded platform are called by CUDA library function to accelerate the tracking speed by the parallel operation.Then,this paper designs the human-computer interaction interface based on Qt to control the airborne tracker and shows the tracking results.Finally,according to the stability and reliability requirements of the system,the function of anti-blocking,scene adaptability,human-computer interaction and the performance parameters such as tracking speed and accuracy of the airborne tracking system are tested.The results show that the airborne target tracker can track the ground targets accurately in both infrared and visible scenes.The system can handle the occlusion problems encountered in the tracking process and support the human-computer interaction of the system.The tracking parameters of the system meet the requirements of airborne tracker.