Automatic target recognition in ultra-wideband 3-D images for concealed weapon detection

Possibility to use computer vision in order to replace human operators in ultra-wideband (UWB) security screening systems is investigated. A target recognition technique for 3-D UWB images is introduced. Spatial spectrum-based features extracted from 3-D UWB images are used for object classification. An artificial neural network to recognize patterns of 3-D spatial frequencies of concealed weapons is developed and trained. The feasibility of developing a practical automated detection algorithm for UWB 3-D imaging systems is demonstrated experimentally.