Based On The Passive Millimeter Wave Image Sequence Of Target Detection And Tracking Algorithm Research

Passive millimeter wave (PMMW) imaging technology uses the difference ofradiant energy in the millimeter wave band between the target and background.Compared with optical imaging、infrared imaging and microwave imaging，thistechnology has some advantages， such as high resolution， high anti-jammingcapability，ability of all-time working and penetrating clouds or textiles，none radiationhazards on human body and so on，so it has become important and new imagingtechnology. Moving target detection and tracking is the foundation and key technologyto achieve the intelligent monitoring in passive millimeter wave imaging system,therefore, it is one of the focus research of the current domestic and foreign scholars.Its main work is detecting the moving objects relative to the whole scene in imagesequences, then, tracking the objects to get their motion state.This dissertation is rely on specific research projects, and do the followingresearch on the moving targets detection and tracking algorithm based on passivemillimeter wave image sequences:1. For fail problem of using direct inter-frame difference method to detect movingtargets under dynamic scenes, introduce a registration algorithm based on mutualinformation to the background registration, and using inter-frame differencemultiplication method to target detection. This algorithm has higher registrationprecision and less residual noise in the difference image, therefore, has better detectionresults.2. For multi-modal problem of scene, research moving target detection algorithmbased on mixture Gaussian modeling method, use multiple Gaussian distributions tomodel each pixel and updated in real time. This algorithm can model to getbackground image closer to the real scene for multi-modal and slow-changing scene.3. For the slow boot problem of moving target detection algorithm based onmixture Gaussian modeling method, present an improved mixture Gaussian modelingmethod. The algorithm removes data of noise and moving target from first N frame image data, then, calculate the mean and variance which are used as initializationparameters, therefore, stable and accurate background image can be quickly got.4. Kalman filter is used to track the detected moving target. Present a method ofweighted objective correlation and a management strategy of delaying to determine thetrajectory. The algorithm can greatly reduce the error rate of target association, besides,it can achieve target tracking while target is missing temporarily.The above algorithms have been verified their effectiveness by experiments. Theresults show, the algorithms based on maximum mutual information and mixtureGaussian modeling method can detect moving targets in passive millimeter waveimage sequences accurately, the improved algorithms boot faster; and the algorithmbased on Kalman filter can track moving targets accurately.