Research On Infrared Target Tracking Technology In Complex Background Conditions

Infrared imaging system adopts passive thermal imaging technology,it has good crypticity and anti-interference,is widely used in security prevention and control,traffic control,military defense and many other fields.Infrared target tracking technology is the key technology of infrared imaging system in various system applications.Compared to visible light images,infrared images have poor imaging quality and low contrast and signalto-noise,in complex application scenarios,it is often influenced by factors such as occlusion,lighting,and interference from similar objects,so improving the stability of infrared targets under complex interference conditions is a difficult problem in current research.Therefore,the research and improvement of infrared target tracking algorithm has great significance in engineering application and theoretical value.This thesis analyzes the technical difficulties of infrared target tracking under complex background conditions,proposes algorithm to solve the problem of infrared target tracking,and proves the effects and performance of the algorithms by experiments.Aiming at problems such as background interference,this thesis starts from the contextaware framework KCF?CA,and explains the problem of too rigid background suppression mechanism and the tracking drift problem common to KCF algorithms in the framework.Based on the results of the cause analysis,this thesis proposes a novel solution that can solve the above problems at the same time to predict the target motion position.It not only enables the background suppression mechanism of KCF?CA to suppress the interference information on the target motion path in advance,but also corrects the position of the correlation filtered sampling center in time to reduce tracking drift effectively.Aiming at the occlusion problem,this thesis verifies that the LBP feature and the HOG feature used in the KCF class algorithm have different sensitivities to the target occlusion situation,but are the same for the rapid change of the target.Moreover,this thesis proposes an occlusion evaluation mechanism based on the LBP feature.This mechanism can distinguish between two situations: target occlusion and rapid target change,and thus adopts reduce and increase the learning rate in this two different strategies.This thesis also designed the occlusion detector to enable the algorithm to adjust the learning rate adaptively better when dealing with occlusion situations.