Infrared Target Detection And Tracking Based On Human Visual Mechanism And Particle Filter

In recent years,infrared target detection and tracking technology has been widely concerned in various countries due to its good concealment and anti-interference ability.Infrared target detection and tracking technology is widely used in infrared guidance,satellite warning,unmanned reconnaissance,industry,agriculture,medicine,transportation,aerospace and other fields.However,there is still a problem that the detection and tracking accuracy of the infrared target in a complex background is low.Therefore,this paper studies the detection and tracking algorithms of infrared targets under complex background.The overall idea of the article is to track before detect,the main research object is the infinitesimal infrared target and the specific conventional infrared target.In order to improve the accuracy and real-time of detection and tracking,the related research and analysis methods about infrared detection technology are provided.Human visual contrast mechanism is used to detect infrared dim and small targets in complex background,the algorithm simulates the mechanism that human eye is sensitive to the target contrast.Firstly,the gradient saliency map of the infrared image is extracted by the8 direction gradient equation and binarization algorithm.According to the size characteristics of small targets,the gradient saliency map is optimized to eliminate the isolated noise points and the larger area of the background gradient.The visual contrast mechanism is used to calculate the local contrast of the optimized saliency map and the false target is eliminated by the threshold processing,which lead to the completion of the infrared dim target detection.In the case of the detection of conventional specific infrared targets,this paper introduces an infrared target detection algorithm based on convolution neural network.The convolution neural network is applied to the detection of infrared specific target.The convolution neural network is trained iteratively using positive and negative infrared target samples and tested successfully.The detection method can effectively detect a specific infrared target.A gravity optimized particle filter algorithm is proposed.The particle swarm is optimized by gravity algorithm in particle filter to improve the filtering accuracy.Each particle is regarded as a mass point and the mass is proportional to particle weight.The gravity attracts particles moving toward high likelihood region which optimizes particle swarm.Then elite particle strategy is introduced to accelerate the particle convergence rate and avoid local optimum in the gravity algorithm.Perceptual model is used to prevent particles from crowding or overlapping due to excessive convergence.When tracking infrared targets,the observation model of multi feature fusion is applied to the particle filter tracking algorithm.The improved gravitation algorithm is applied to the target tracking process.Considering the limitation of infrared target feature description,We build the characteristic observation model based on the histogram of gray and the histograms of oriented gradients.The particle weights are updated by calculating the Bhattacharyya distance between the candidate template and the target template.In order to improve the accuracy of target tracking,an adaptive dynamic updating strategy is adopted for the target template.