Research On Infrared Small Target Detection And Tracking Algorithms

Infrared imaging technology has many advantages such as well-concealment and strong anti-interference ability,it gets a widely use in the military and civilian applications fields.In practical applications,because of the long imaging distance,the infrared targets are small and lack of shape and texture features,thus making target detection and tracking very difficult.Therefore,achieving accurate detection and tracking of infrared small targets is an important and difficult task.In this thesis,it is proposed by using Tophat operator and improved Robinson guard filter to suppress background of infrared images,and using adaptive threshold segmentation to extract the candidate infrared small target regions.On the basis of extracting the candidate infrared small target regions,this thesis proposes two different approaches to detect infrared small targets,one is multi-filters fusion infrared small target detection algorithm based on traditional filtering ideas,the true infrared small targets are extracted from the candidate region by combining the Unger smoothing filter with the imaging uniqueness of the infrared small targets.The other is ITNet(Infrared Target Network)infrared small target detection algorithm based on the deep learning convolution neural network(CNN),and using ITNet networks to identify true infrared targets from candidate targets.When infrared video sequences are processing,using multi-object association filter to further remove the pseudo-infrared small target,lower the false alarm rate.Based on the idea of pipeline filtering,this thesis proposes a multi-objective data association tracking algorithm with small target detection results,and establishes multi-objective data association matrix for multi-objective state analysis to complete multi-object tracking task.When the small targets disappear or target dimming,the motion parameters of the small targets are predicted based on the particle filter algorithm.The local area of the dim targets is enhanced by the Retinex algorithm to determine whether the dim targets exist.The experimental results show that the algorithm proposed in this thesis achieved high detection accuracy under different background images,even when the target is submerged in the background,it can be used to analyze the performance of the algorithm.