Target Tracking And Autonomous Landing Of A Quadrotor Based On Vision

With the rapid development of artificial intelligence technology,the quadrotor is widely applied in people's production and life,such as military navigation,security monitoring and other fields.But now,most of the aircraft need to control the flight via the remote control and have limited endurance.How to make the quadrotor perform target tracking task on the basis of autonomous flight and achieve autonomous landing when it needs to return has always been a research hotspot and difficulty in the field of mobile robots.In this paper,several key problems of target tracking for quadrotor in a 3D scene are studied,including target scale change,target occlusion and target shake.Aiming at the target scale change problem,this paper compensates the depth information for the 2D image sequence by joint calibration of the monocular vision sensor and the binocular vision sensor,and realizes the mutual conversion of the absolute scale and the relative scale.And it makes the Kernel Correlation Filter better adapt to the scale variation of the target.For the target occlusion problem,this paper first analyzes that the target occlusion is due to the loss of depth information in the process of mapping a 3D scene to a 2D image,so we proposed a method of hierarchical expression of scene information.Based on this method and Kalman filter,the occlusion behavior can be predicted.At the same time,the modified Artificial Potential Fields method is used to adjust the motion trajectory of the quadrotor in real time to reduce the occurrence of occlusion.Aiming at the problem of the sudden change and blur of tracked target position caused by aircraft shake,this paper uses Kalman filter to correct the target sampling area,which improves the accuracy of the target tracking algorithm to some extent.When the quadrotor is required to return for some reason(such as poor battery life),the autonomous landing based on visual aid signs can achieve safe and accurate landing.In view of the problem that most visual signs can not provide multi-scale detection,we designed a visual aid sign based on concentric circles,and achieved the target detection and precise positioning of this visual sign.In addition,we also designed the PTZ controller,speed controller and position controller.Finally,the quadrotor successfully landed on the roof of the unmanned vehicle with unknown speed.In this paper,DJI M100 quadrotor made by DJI was used as the experimental platform,and the effectiveness and practicability of the algorithms mentioned above were verified through experiments.