SBIR-STTR Award

This Phase I proposal describes the development of an Automatic Threat Recognition (ATR) algorithm for volumetric CT data. Our ATR algorithm consists of four stages: (1) preprocessing of CT data, (2) object segmentation of preprocessed CT data, (3) post-processing of segmentation results, and (4) explosive detection from the segmented objects. The ATR algorithm will be made computationally efficient by GPGPU programming in order to meet desired throughput of the screening process. The performance of the ATR algorithm will be thoroughly analyzed via extensive experiments with CT dataset of typical checked bags with ground truth. Since a preliminary segmentation algorithm is already built, Phase I begins from TRL 3 and concludes with experiments and comprehensive quantitative performance analysis (TRL 4). A DICOS-compliant ATR algorithm and standard CT test datasets for reliable quantification of explosive detection performance is expected at the end of Phase II.

For Phase II work, we propose to complete the development of the DICOS compliant ATR algorithm whose feasibility has been demonstrated by results of Phase I work. Important project milestones for the first base year of Phase II are as follows: (1) implementation of the DICOS standard including network protocol, (2) improvement of the proposed ATR algorithm in terms of explosive detection performance and computational efficiency, (3) development of plug-in ATR software with the DICOS standard, and (4) optimal parameter tuning using the receiver operating characteristics analysis. The goal of the second option year of Phase II is to be ready for the certification test by continuing improving the ATR algorithm and optimizing parameters via extensive experiments with training datasets. With the successful completion of Phase II work, we expect that our ATR algorithm will outperform existing ATR algorithms of EDS vendors in terms of PD and PFA while meeting the certification requirement of throughput. If our ATR algorithm is certified by TSL and used for EDS, it will contribute to enhance aviation security significantly. Since our ATR algorithm will be able to lower PFA while maintaining PD of the current state-of-the-art EDS, it will be also able to reduce the cost related to manual inspection of alarmed bags.