Research On Stable Tracking Algorithm For Moving Target

Video-based target tracking technology is a technical means of locking a specific target in a video stream.It combines the leading technologies of computer vision,digital image processing,pattern recognition,artificial intelligence and automatic control,and it has a wide range of applications in military combat,defense,security monitoring,intelligent transportation and other fields.In recent years,it has received extensive attention from scholars.The target tracking algorithm based on correlation filtering(CF,Correlation Filtering)is simple and flexible.It has adaptability to the affine transformation,occlusion,complex background of the target,and can better balance the tracking accuracy and speed.This thesis adopts the kernel correlation filtering algorithm(KCF)to construct the basic tracking framework.In order to achieve the automatic recognition of the target before tracking and the re-detection after the target tracking failure,the YOLO(you only look once)target detection algorithm has been studied and merged with KCF.The research content of this thesis mainly includes the following points:1.We have studied the framework and mathematical principles of the correlation filtering algorithms,and briefly described their follow-up research and development status.2.we have studied the classic algorithm: kernel correlation filtering(KCF),which is based on CF framework.In order to solve the problem of insufficient spatial resolution of Fhog feature,a frequency domain interpolation method is proposed,which improve the KCF's “sawtooth” when tracking moving objects.And based on the scale pool method proposed by SAMF,the multi-scale tracking extension of KCF is carried out.3.In order to achieve the automatic recognition of the target before tracking and the re-detection after the target tracking failure,The YOLO target detection algorithm was studied.In addition,we produced a quadrotor UAV dataset and trained the UAV target detection model accordingly.4.The detection and tracking algorithm is integrated to achieve automatic capture of the target and re-detection during the tracking process.And deployed on Jetson TX2 platform.5.In this thesis,all the algorithms are experimentally verified and all source code is provided.