Dual-view digital radiography for luggage inspection
University of New Brunswick
This project examined the use of two orthogonal digital radiographs to detect the presence and location of any contraband that may be concealed in a luggage or a pallet-sized cargo container. Detected radiographic projection intensities were divided into distinct intervals, which were paired, one from each radiograph, and backprojected, to form a "local" image. Each local image was used to estimate the width of apparent articles along the two directions of radiation incidence. The recorded projections were normalized with the estimated width, and introduced into a network flow algorithm as supply sources in one radiograph and demand sinks in the other. Designating the material to be detected as a "target" material, a cost matrix was calculated. Each element in this global matrix corresponded to the location of a pixel in the object's image, and the value of the matrix's element represented the probability of the corresponding pixel being occupied by the target material. This enabled the segregation of target materials from the surroundings by thresholding the matrix values with a cut-off quantity (minimal cost). This approach was applied to target materials of various densities to determine the presence, size and location of each of them. The target-based method was tested in simulations of simple configurations, a luggage-like setup, and a pallet-sized cargo arrangement, demonstrating its ability to detect suspicious articles in most configurations, with a smaller number of false indications than in an approach based only on the intensity of paired projections. This method needs, however, further laboratory testing before using in field applications.