Ground Moving Target Monitoring Based On UAV

Unmanned aerial vehicles(UAVs)have developed rapidly in the military and civilian fields in recent years for the advantages of flexible,remote controllable,unaffected by the ground traffic congestion and have a large field of view(FOV).UAVs are equipped with various types of visual sensors for ground moving target monitoring.They can be widely used in battlefield reconnaissance,traffic monitoring,disaster inspections,and other tasks.Therefore,they have become a research hotspot in the field of UAVs.In current target monitoring methods,most UAVs are equipped with auxiliary and active sensors,such as laser rangefinder.However,these active sensors are no longer applicable in complex environments such as cities and battlefields.This paper focuses on the target monitoring methods from UAV using monocular camera and inertial navigation sensors.The main research contents and contributions are as follows:In view of the limitations of on-board processor and the difficulty of detecting and tracking small ground moving targets in the complex scenes,we analyzed the state-ofthe-art algorithms,and designed an effective framework to detect and track the moving target in real time.An aerial dataset containing small targets was collected to verify the effectiveness of the methods.Experiments show that the m AP of the small target detection methods is 90.61% based on our dataset,and the tracking method is robust and high precision.The interested target could be detected and tracked in real time.The moving target is difficult to geo-locate for the limited performance of on-board sensors and monocular camera.Aiming at solving these problems,a passive target geolocation method from UAV using monocular camera is developed to obtain the location and direction of the ground moving target.The proposed scheme has been built on a UAV platform on which a monocular camera and a microcomputer are added.Experiments show that the error of this passive geolocation method is 5m when the flight height of the UAV is about 100 meters.The attitude angles of gimbal camera are changed robustly to keep the target in the FOV of UAV,which is the result of the Indirect direction calculation algorithm.Considering the limitation of the observation range from a single UAV,a distributed fusion method based on Interacting Multiple Model(IMM)and Unscented Information Filter(UIF)is proposed to estimate the motion of ground target.This method overcomes obstacles such as high mobility of target,high non-linearity of the observation function,communication bandwidth limitation,and distributed system consensus.It also improves the accuracy of target motion estimation.Simulation experiments show that this method can quickly converge.The Root Mean Square Error(RMSE)of position is 2m,and the RMSE of velocity is 1m/s.The average consensus error of position and velocity is 0.5m and 0.2 m/s respectively.In this paper,we designed a scheme which is capable of detecting,tracking and geo-locating the ground moving target from single UAV platform,and proposed a distributed fusion estimation method of target motion based on multi-UAV platform.This framework has been served for the National Key Research and Development Program of China "Urban Fine Scene Space-Time Awareness Equipment and Online Monitoring Technology".