Design And Implementation Of Vision Based Autonomous Target Tracking System For Quadrotor

With the continuous increase of computing power of chips,intelligent mobile robot systems are more and more widely used.Especially the quadrotor UAV system has penetrated into the film production and consumer-grade aerial photography,surveying,electric power patrol and even military reconnaissance fields.The capability of detecting and tracking interested targets is of great significance to the quadrotor UAV system.This paper mainly focuses on the design and implementation of a complete and practical vision-based quadrotor target tracking system.To satisfiy with the actual engineering requirements,the advanced vision tracking algorithm is improved.The state estimation and tracking control of the tracking target are designed and implemented.Finally,the quadrotor target tracking system including the ground control station,quadrotor and a complete set hardware and software is completed.The reliability and practicability of the system are verified by simulation and actual flight test.The main work and contributions of this paper are as follows.1)According to the actual needs,the selection of onboard computers,visual sensors,power components and so on has been completed,and a quadrotor UAV platform has been designed and built.The overall software design scheme of the quadrotor target tracking system has been designed in detail.2)Modeling of quadrotor target tracking system is carried out,including quadrotor dynamic model,visual imaging model,definition of relevant coordinate system involved in navigation and guidance.Besides,kaliber calibration experiment is carried out to obtain the transformation relationship between coordinate systems.Finally,the models for simulating visual target tracking simulation experiment is completed under the gazebo simulation environment.3)Several advanced visual tracking algorithms are studied and compared.Based on the short-term visual tracking algorithm FDSST published on TPAMI 2017,a robust and real-time long-term tracking algorithm is obtained by adding appearance filter and SVM classifier as the long-term tracking strategy.4)The traditional L-K optical flow method is improved to calculate binocular parallax and get the original measurement value of relative distance between the tracking target and UAV according to the imaging model.Then the Kalman filter is used to estimate the global position and velocity of the tracking target.Finally,a PID controller is designed to realize the UAV's stable tracking of the target.5)For the proposed visual tracking algorithm,this paper compares and evaluates it with the state-of-the-art tracking algorithm on OTB data sets and video sequences from our quadrotor's onboard camera.It verifies the advantages of the proposed algorithm in terms of long-term tracking ability,illumination variation robustness,low-resolution targets robustness,as well as overall success rate and real-time performance of the algorithm.The effectiveness of the target state estimation algorithm and the reliability of the whole quadrotor target tracking system are verified by simulation and actual flight experiments.