Deep Learning Algorithms for On-Person Stream of Commerce Characterization

Synthetik proposes to develop a suite of deep learning algorithms for analysis and characterization of on-person stream of commerce at airport security checkpoints. The algorithms will be optimized for real-time processing on existing CCTV video streams at pre-checkpoint zones and deployed on commercial off-the-shelf hardware to minimize hardware development costs and protect personally identifiable information (PII). The proposed algorithm development will leverage prior work performed by Synthetik on synthetic data generation for checkpoint passenger screening systems. The on-person stream of commerce analysis tool will calculate statistics on the distribution of observed anthropometric characteristics and clothing articles that will inform the Government and commercial vendors of walk-through screening systems, thereby improving algorithm performance, TSA operational costs, and passenger experience. The Phase I effort will focus on developing passenger tracking, anthropometric measurement, and clothing classification algorithms to demonstrate feasibility of the proposed tool. An analysis and optimization of these algorithms will be performed to identify software and hardware requirements for the prototype, which will be developed and deployed in a representative operational environment during Phase II.