Quad-rotor Aircraft Target Recognition And Tracking Based On Vision

Nowadays, with the widely use of Unmanned Aerial Vehicles (UAV) in the warface, the searchand investigation of the intelligence is more and more significant. The automatic target recognitionand tracking technology based on image is an important problem needed to be resolved in the areas ofUAV navigation based on c and a focus in the field of computer vision.This paper, whoseexperimental platform is the quad-rotor aircraft, mainly discusses the problem of target recognitionand tracking based on vision, complete a set of processes of system design, identification and trackingalgorithm design and experimental validation.Firstly, according to the structure and flight mode of the quad-rotor aircraft, a design for thetracking system based on vision is proposed. Then the paper introduces the hardwares of the system,especially the visual module, and summarizes the workflow of the system.Secondly, in order to recognize the target by image-matching, extract SIFT feature after imagepre-processing, then propose a PCA-KD query algorithm to improve the matching accuracy, andpropose a RANSAC algorithm to take a further purification of the mathing results. The experimentresults show that the advanced algorithm can recognize the target even when the target scale changesor partial blocked.Thirdly, after analyzing the advantages and disadvantages of two tracking algorithm: Camshiftand Particle filter, an advanced algorithm which absorbs the advantages is proposed.First determinethe intitial position and size of the tracking window, then combine the color feature, edge feature,SIFT feature, propose the fusion strategy and workflow of the tracking algorithm or optimize thespread of particles by Camshift. In the end, the experiment shows that the tracking algorithm isavailable either among similar colors or in obscured circumstances.Finally, experiments of quad-rotor aircraft flight tracking are conducted. First introduce thepinhole imaging model, then develop a stable vehicle tracking solution and propose the switchingprocess of tracking status. The experiment results show that the system has good capability of targetrecognition and tracking, prove that the algorithm in this paper is available and accurate.