Human Body Hidden Contraband In Passive Millimeter Wave Image Recognition Research

In view of the current situation that passive millimeter wave image quality for body concealed weapons is poor and target detection is based on people’s subjective judgment, a method on concealed weapons detection automatically and identification in millimeter wave image was presented in this paper. A cascade classifier was constructed by using the Haar-like features combined with AdaBoost algorithm to detect the weapons, and a concealed weapons Hidden Markov Model database was established to recognize the target.The original millimeter wave image has low resolution and is with noise. An estimating noise’s type method based on wavelet domain was presented in the paper, and the main Gaussian noise was removed by wiener filtering.A method of image segmentation based on the EM algorithm was adopted, and the regions where gray value close to the target area were split out, which offered samples for classifier design and training Hidden Markov Model. Gun classifier was designed by training Haar feature classifier based on the AdaBoost algorithm. A classifier which detection results conformed to the system requirements was achieved by using different samples, lowering cascade classifier series, and lowering requirements of the false-positive rate. In this paper, gun Hidden Markov Model database was established respectively by using two dimension discrete cosine coefficient, singular value decomposition, invariant moment and shape descriptor as feature extraction method. After the analysis and comparison, the gun Hidden Markov Model which features extraction was combining Fourier descriptor with discrete cosine descriptor could identify guns in different proportion and angle. Experimental results showed that the pistol HMM has invariance to rotation, translation and scale.An algorithm on automatic detection and recognition for body concealed weapon based on Haar-like feature, AdaBoost and HMM was presented. Experimental results demonstrated that the algorithm could accurately detect and identify the pistol hidden in the clothes in millimeter-wave image.