Research On Targets Tracking Algorithm Of Passive Millimeter Wave Image And Video Sequences

A passive millimeter wave (PMMW) imaging system achieves images through thedifference of intensity that scenes and objects radiate, which has great performance inpenetrating the clothes and other things. What’s more, it has an excellent ability ofrecognizing a metal target from its surrounding environment. However, because of thediffraction limit of the antenna, the spatial resolution and Signal Noise Ratio (SNR) ofPMMW image is relatively low, leading to the lack of detailed description of the target’sshape and texture. Since the target tracking algorithm performance of PMMW dependson the information of image, these features may lead to the lost of target or themiscarriage of justice. Using the tracking algorithm based on the fusion of PMMW andvisible images can solve this problem. It can not only extract the shape and texture oftargets through the visible image, but also probe the metal targets information throughthe PMMW image, which is benefit to the positioning and tracking of targets.Based on the actually specific research projects, this article studies the trackingalgorithm aimed at the above problems. The mainly work include:1. The centralized/distributed/hybrid model of the multi-sensor data fusion isanalyzed. Through the comparison their advantages, disadvantages and the applicationenvironment, the viewpoint that the distributed fusion model can reduce the error offusion system and can improve the real-time of the system is summarized.2. Based on the imaging characteristics of PMMW image and video sequenceimage, the feature of targets in these two kinds of image is analyzed. The feature levelfusion method of the targets’ feature such as position and areas is researched, which canimprove the information utilization of the different sensor image.3. Aiming at the problem that the traditional linear target tracking algorithm has alarge error, the radial basis neural network is introduced in the state estimation of targets.A method that can reduce the mean value of the error and the times of learning by usingthe error factor weight is proposed, which is more suitable for the tracking filteringalgorithm with high robustness requirement.4. Aiming at the problem that the position of targets that extracted from PMMW image and visible image are quite different, the targets matching probability fusionalgorithm is proposed. Through calculating the probability of the targets matching area,the targets are classified according the judgment, and the fusion result can be got, whichis benefit to the targets tracking.