Infrared Moving Object Detection Based On Gaussian Mixture Model

Moving Target detection is the foundational research in the computer vision, artificialintelligence, pattern recognition and other fields. Its essence is to separate the target from thebackground accurately based on the motion characteristics of the target. The accuracy of thetest results has important influence on subsequent target classification, tracking,identification, etc. Moving target detection based on infrared image sequences can be doneat night, in the rain, snow, fogs and other low visual environment, the detection has beenwidely used in the field of military and civil application.Now, even if the advanced infrared imaging equipment still cannot get rid of theinterference of random noise, and choosing a reasonable method to suppress or eliminate thenoise has a positive effect on the accuracy of target detection. This paper introduces severalalgorithms for image denoising, then presents an infrared image denoising algorithm basedon the Bidimensional Empirical Mode Decomposition and threshold estimation with allof sub-bands. First, decomposing the noise image through the Bidimensional EmpiricalMode Decomposition to obtain several sub-bands’ information. Second, calculating thesub-bands’ noise variance to get a reasonable threshold for noise removal. Finally useinverse transform to reconstruct image. The experimental results show that the qualityof theimage was improved both in the objective evaluation indexes and in the overall vision.Infrared images sequence with low resolution lack texture and detail featureinformation, how to detect the moving targets completely and accurately has been thedifficult focus of the current research. Using the traditional Gaussian Mixture Model candetect general position of moving targets, but also has some obvious shortcomings: such as the model update rate is fixed, detecting target with false contours, interior is full of holesand even the overall fracture. In order to solve the above problems, firstly this paper usedStructural Similarity Algorithm to divide mage into different regions, each area usesdifferent rates to update the Gaussian Mixture Model. Secondly, in order to obtain moreaccurate motion target, the Gaussian Mixture Model is used to locate the infrared target areaand the Watershed Algorithm based on spatial information to make target area closed.Finally, use Pulse Coupled Neural Network Algorithm to segment the closed areas andextract moving target from the background effectively. The method can avoid somedisadvantages in traditional methods and get more complete and accurate infrared movingtargets.To sum up, this paper has a in-depth research on the infrared image denoising anddetecting of moving target based on infrared image sequences, Put forward some new ideasand methods which were proved to get a good image processing result, Provides referencefor the research and development of the infrared imaging technology.