Design And Implementation Of Ground Target Tracking System Based On Multi-rotor UAV

In recent years,air-ground collaborative robots composed of a multi-rotor UAV and a land robot have been widely used in emergency rescue,detection and disposal scenarios.In order to realize the target following and precise landing functions of the multi-rotor UAV,the visual target detection and tracking method suitable for the UAV is deeply studied,and the ground target tracking system based on multi-rotor UAV is designed and implemented.The specific research work of the paper is as follows:(1)Aiming at the problem that the deep learning target detection method is difficult to run in real time on the onboard processor,a positioning label deployed on the land robot and its detection method are proposed.The label is detected by color and outline features,which is used to obtain the position of the land robot in the image plane.Experiments show that the average frame rate of the method on the onboard processor is 51 frames per second,which can meet the requirements for real-time running.(2)Aiming at the low tracking accuracy and robustness of the current mainstream Kernelized Correlation Filters method when the image scale change,perspective change and the target beyond the field of view,the tracking accuracy of the method in the scene of scale change is improved by finding the maximum response of the filter at different scales,and the Kalman Filter method is used to predict the search area,which solves the problem of tracking failure caused by the target beyond the search area due to the large perspective change.Finally,the April Tag visual positioning system is introduced to automatically select the target,and the effectiveness of the tracking results is judged according to the filter response value.When the tracking fails,April Tag is used to redetect the target,which solves the problems of inaccurate manual selection of the target and tracking failure when the target beyond the field of view.Experiments show that the average pixel error of the improved method is 6.8 pixels and the average overlap rate is 0.67 in the scene of scale change,which is better than the 11.7 pixels and 0.42 before the improvement.Besides,the improved method can cope with the perspective change and the target beyond the field of view.(3)In order to realize the target following and precise landing function of the multirotor UAV,the target position in the world coordinate system is solved by analyzing the transformation relationships between the coordinate systems of the system.The desired position of the UAV is planned according to the target position,and the position and attitude of the UAV are controlled in combination with the fixed-point control model.Finally,the ground target tracking system based on the multi-rotor UAV is built,and the indoor and outdoor target tracking and precision landing experiments are carried out.Experiments shows that the following and landing error of the system is about 0.1m,which has certain engineering significance and application value.