Drone-based Moving Target Surveillance In Dense-obstructer Environment

Moving target surveillance is a basic function of drone-based systems,which could give rise to various potential applications in smart city and intelligent transportation.However,how to monitor moving targets in the dynamic high-altitude environment with restricted computational resources is one of the most critical challenges to be solved.Furthermore,during the surveillance,the moving target could travel in dynamic speed and be blocked by dense-obstructer,such as woods and buildings.Hence,traditional monitoring systems are not suitable for those applications with problems mentioned above.Recently,due to the fast development in the drone and sensor technologies,drone-based moving target surveillance systems have received more and more attention.This paper studies the online moving target surveillance technology based on the drone.Our work is as follows:(1)In the view of limited drone energy,we design an energy cost function-based adaptive path planning method for the drone.It first generates a digraph based on the Voronoi diagram,then calculates energy loss of adjacent waypoints via the wind field and the distance to the starting point,and finally determines the most suitable flight path by evaluating the energy cost.(2)To detect moving targets from the dynamic background,we design a online moving target detection algorithm based on SURF feature points.First,it extracts the SURF feature points from the image;second,it uses the ANN algorithm to match the background feature points;third,it calibrates images through the perspective transformation matrix;finally,it utilizes the SRI method to segment the foreground targets from the image.The algorithm does not need any prior knowledge of the target.The experimental results show that our algorithm has high target detection accuracy in the dense-obstructer environment,and can meet real-time requirement.(3)For dense obstructers in the environment,we design a online moving target tracking algorithm based on kernel correlation filter model.It firstly establishes a tracking model based on the target detection results and the image information of the current frame,then trains a correlation filter to predict the position of the target in the next frame.Also,it fixes the problem of missing target through the information from the detection module.Additionally,for the problem of target deformation and unknown motion model,we integrate SURF features and HOG features to enhance the robustness of target tracking.Finally,we design and realize the online drone-based moving target surveillance system,and carry out the experiments under the real environment.Experimental results show that the system can detect and track the moving target in real time with high accuracy.