Ground Target Recognition And Tracking Based On Airborne Vision

With rapid developments of unmanned aerial vehicle(UAV)technology,researches on the applications of UAV have been deepened and the functionalities of UAV have also been greatly enhanced.Among which,UAV airborne GPS,inertial navigation,radar,machine vision and other sensing devices play a key role,providing not only guarantees for their UAV flight control,but also significant means for UAV target monitoring,tracking and Distance-Measurement and other tasks.Ground target recognition and tracking,based on airborne vision,is one of the important functions of UAV applications,which can be used in military reconnaissance,urban monitoring,intelligent transportation and other fields.With its high resistance to temperature,magnetic field and other interferences,vision sensors can obtain richer information from the outside world and reflect the target position and movement more accurately.In this paper,an airborne vision system was built on a quadrotor,and an algorithm of recognizing and tracking ground targets based on machine vision was studied,which could solve the problem of ground target recognition and tracking in complex background perfectly.In this paper,the moving vehicle was chosen as the ground target,characteristics of the target color,shape and other main characteristics were analysed,the hardware structure and image information processing way of the airborne vision system were designed,and the overall scheme of the ground target recognition and tracking technology research was set up.When tackling the problem of ground target recognition,a method,which combines global features and local features,has been adopted by this paper to realize the recognition process.Firstly,the region of the ground target was preliminarily determined according to the color feature and the edge feature,and then the SURF feature was used to locate the target accurately.The method effectively reduced the interference of other objects with the same color or shape to the target recognition in the background,and improved the accuracy and robustness of the algorithm.As for the interferences from multi-moving objects and occlusions in the background,during tracking the target,a method adopted multi-feature fusion CamShift algorithm,which is based on Kalman filtering,has been put forward.The proposed method could improve the traditional CamShift tracking algorithm by weighted fusion of the color histogram feature and the edge orientation histogram feature,forecast the actual state of the target via Kalman filter,and finally accomplish the accurate tracking of the ground target.Furthermore,the ground target recognition experiment under complex background conditions and tracking experiment under similar color background and occlusion conditions were completed.The experimental results show that the recognition rate is above 95%,the average recognition time is 283 ms,and the average tracking time per frame is 159 ms.It is convinced that the target recognition and tracking algorithm proposed in this paper could realize ground target recognition and tracking under complex scenes quickly and accurately,which not only solves the UAV ground target tracking problem,but also provides reference for other forms of high-altitude video monitoring.