Research And Achievement On Moving Targets Tracking Algorithm Of Passive Millimeter Wave Imaging

The passive millimeter wave(PMMW) detection system can detect the energy of millimeter wave band that come from environmental radiation to achieve imaging. Millimeter band waves can penetrate the dust, clothing, rain and fog, etc. to spread, it has the characteristics of imaging the occult target without the influence of the climate. So the use of millimeter wave passive detection system for imaging the hidden target in the scenes and use target tracking technology to track the conceal target has greater significance and value. However, the low resolution and noisy of the passive millimeter wave image which is effected by imaging device brings a certain degree of difficulty for the study of the passive millimeter wave image tracking technology.This paper aims to study passive millimeter wave image tracking technology, the main contents and conclusions include the following aspects:(1)Study the theory of passive millimeter wave imaging and the working principle of passive millimeter wave detection system. Then, confirmed the targeted and relevant ideas of this thesis.(2)According to the characteristics of the low resolution of the image caused by actual project, we studied several gray features of the image and compressive tracking algorithm. For the tracking drift of compressive tracking algorithm when the targets’ posture changed in passive millimeter wave image and the obstruction sheltered the target, we proposed a compressive tracking algorithm based on Histograms of Oriented Gradients Feature and an adaptive update rate compressive tracking algorithm based on Scale-invariant feature transform feature, solving the problem of inaccurate description of the target and the changeless of the update rate in compressive tracking algorithm. The experimental results show the algorithms have better performance than the compressive tracking algorithm.(3)According to the characteristics of the low SNR of the image caused by actual project, we studied the integrated learning, which has a strong generalization. For the problem of sensitivity to noise due to the update rate and the samples could not be obtained in compressive tracking algorithm, we studied the online learning and proposed an improved fast compressive tracking algorithm based on online random forest classifier,solving the tracking drift because of the stripe noise of image, the occlusion of the obstruction and the change posture of the targets. The experimental results show that the algorithm tracked the target more precisely.(4)According to the similar gray distribution between the background and target security environment, the different gray distribution between target and human body, we proposed a target detection algorithm based on the detection of human body. For the problem of disappear of the target and the incomplete accuracy of the tracking algorithm, we proposed the detection while tracking algorithm, solving the problem of false alarm and undetected of the target. The experimental results show that the algorithm can achieve fast and stable tracking.