Research On UAV Target Following Technology Based On Computer Vision

In the traditional following mode,UAV implements the follow function by following a signal source.In this mode,UAV does not know what the specific target is,and because the target needs to carry a signal source,UAV cannot follow it quietly,and this mode does not work well when the source of the signal is interfered by other signals.In response to those problems,this paper applies computer vision to UAV working outdoors,which solves the shortcomings in the traditional follow mode,and uses computer vision to detect and acquire specific targets,so as to achieve real-time tracking.According to the needs of UAV target following,this paper chooses a relatively novel target detection algorithm Yolov4.On this basis,the non-maximum suppression step in the original algorithm is optimized,and according to the characteristics of the UAV following the target in the experiment,The detection loss function in the target recognition algorithm is optimized,and an improved image recognition algorithm—Single-Yolov4 is proposed.This algorithm speeds up the training process of the target model and improves the training speed of the model effectively.In the aspect of UAV following,the method is to record the moving path of the target on the image,and convert the moving distance of the target on the image to the real world through the actual relationship between the image and the object.UAVs accordingly direction and distance to move themselves.At the end of this paper,under linux system,gazebo and ros are used to verify the feasibility of this algorithm through 3D simulation modeling.The experimental results show that the UAV has a good following effect when following low-speed moving targets.When the target moves too fast and runs out of the UAV’s camera field of view,the UAV cannot continue to follow the object.