Research On Intelligent Tracking Method Of UAV Video Object

Object tracking from the perspective of UAV is a key technology for drone reconnaissance,aerial photography,traffic monitoring and power inspections.Multi rotor UAV is light and easy to control,with the advantages of vertical take-off and landing,stable performance and free hover,which can perform the above tasks in complex environment.However,the load and battery capacity of the multi-rotors UAV limit the types of computing devices allowed on the embedded system,which seriously limits the available computing power.In addition,object tracking usually needs to label the initial position of the target manually.In the labeling process,the UAV and target position may have changed,which leads to the failure of tracking.In order to settle these problems,this paper studies the real-time object tracking algorithm which can be used for light embedded devices,and uses saliency detection instead of manual acquisition of general target positioning in the first frame,then builds a quadrotor UAV visual target tracking system to complete the autonomous tracking task.The main work and contributions of this thesis are as follows:At the very beginning,this thesis study the object tracking algorithm based on correlation-filter.Fruther,the formula derivation of Kernel Correlation Filter algorithm in training and detection phase is given.Several improvements are made on the basis of the KCF algorithm: integrate two KCF trackers which are alternately operating by using Kalman filter to estimate the movement of the object to smooth the intermediate output;fusion of multiple features;modification template update strategy.Then it was verified on the OTB100 and UAV123 dataset,the accuracy and success rate plot results show that the improvements are effective.At the same time,the speed of the algorithm still retains.Secondly,the salient object detection algorithm based on minimum barrier distance and its application in object tracking are studied.Introduce the concept of distance map and the loss function of barrier distance.Then the rapid realization of the approximation of barrier scanning algorithm is given.Combined with the image post-processing manual method,the salient object detection is realized.Combined with image processing method,salient object detection is realized and integrated into the process of tracking algorithm.In order to evaluation the effect,the fused algorithm is verified on a partial sequence of the selected dataset and can achieve better tracking results.Finally,the thesis build a visual quadrotor UAV platform based on Raspberry Pi.Design of a UAV’s Rear Tracking Speed Servo Controller Based on the Results of Visual Target Tracking.Subsequently,the simulation tracking experiment is carried out.At last,the stability of the Quadrotor UAV system for autonomous tracking of object is verified.