The Department of Homeland Security Science and Technology Directorate and the Transportation Security Administration announced eight winners who received a combined $1.5 million in its Passenger Screening Algorithm Challenge July 9.
The challenge was designed to look beyond typical sources to garner algorithms that would help improve the speed and efficiency of airport screening processes.
“By reaching beyond the screening equipment industry, we have an opportunity to discover new, nontraditional performers that might otherwise be overlooked,” said William N. Bryan, the DHS senior official performing the duties of under secretary for science and technology.
“Working with algorithm developers to improve screening technologies directly serves S&T’s mission to deliver effective and innovative insight, methods and solutions for the critical needs of the homeland security enterprise.”
Competitors were provided with 1,000 volunteered passenger images to train a threat-detection algorithm, which was then evaluated against a larger dataset of images.
“The Passenger Screening Algorithm Challenge was an innovative way to challenge a broad community to solve a difficult problem,” said Dr. John Fortune, DHS S&T apex screening at speed program manager.
“Better ATR algorithms directly contribute to the passenger experience, reducing the need for pat-downs and accelerating the screening process. Through the prize competition, we’re getting better results than we ordinarily might see, by connecting with very smart people who have great ideas but might not typically be part of a government proposal process.”
The challenge awarded cash prizes to the top eight entries:
- Jeremy Walthers of Rockville, Maryland — created an array of deep-learning models to process screening images from multiple views ($500,000).
- Sergei Fotin of Nashua, New Hampshire — fused 2-D and 3-D data sources to make object location predictions ($300,000).
- David Odaibo and Thomas Anthony of Alabaster, Alabama — used specialized image-level annotations to train their two-stage identification models ($200,000).
- Zach Teed of Hudson, Ohio — used a location-based model to define threats ($100,000).
- Oleg Trott of SanDiego, California — fused 2-D and 3-D data sources with automated image augmentation to improve model accuracy ($100,000).
- Halla Yang of Wilmette, Illinois, and Phillip Adkins of Chicago, Illinois — designed an approach that automatically segmented the image before running models trained on specific cropped images ($100,000).
- Suchir Balaji of Sunnyvale, California — used synthetic data and cross-image analysis to produce more robust predictions ($100,000).
- Michael Avendi of Irvine, California — used separately trained models and random image augmentation for best results ($100,000).
The TSA now plans to test the eight winning algorithms against additional datasets.
“We hope to continue to engage the combined expertise of our winners to improve the winning solutions and ultimately see them used in airport security checkpoints,” said Fortune.
Jessie Bur covers the federal workforce and the changes most likely to impact government employees.