The U.S. Marshals Service, charged with moving prisoners around the nation, is preparing to build a facial recognition tool into its agents’ smartphones to help with speed and accuracy of identifying prisoners during transfer.
The agency released a request for quotes seeking a software development kit, or SDK, that will give its developers a code base to begin constructing a tool that will be integrated with separate, existing in-house apps built by the Marshals Service and the Bureau of Prisons.
The final product will interface with multiple existing systems and allow marshals to take a picture of a prisoner, match that against a preestablished image gallery and get a full rundown of relevant information, including origin and destination, medical information, security concerns and, of course, confirming the prisoner’s identity.
The service is in the midst of digitizing all prisoner information for transfers, such as photos, biographic information, security and medical documents. Currently, this information is combined into a paper packet that accompanies the prisoner during transfers.
In the near future, the Marshals Service’s Justice Prisoner and Alien Transportation System, or JPATS, and the Bureau of Prisons will complete development of the Movement Packet, or MPAC, app, which will send all of the relevant prisoner information digitally to the JPATS Management Information System, or JMIS, app at the time of transfer.
The MPAC app is being developed for the Android operating system, prompting the JPATS to require both Android and iOS compatibility in the facial biometric solicitation.
The JMIS app sits on marshals’ agency-provided iOS devices. The app is used to coordinate every part of a prisoner transfer, including “flight/trip schedule, prisoner manifest, prisoner photo and information, and highlights significant medical, security/intelligence information, as well as provides the ability to add mission comments, incidents and record mission times,” according to the solicitation. The app also features the ability to download this information for ease of access in low-bandwidth areas.
Now, the JPATS team is looking for a SDK to help add facial recognition capabilities to the apps, opting to buy code to use as a base for internal development work rather than buying a fully developed tool and bolting it on.
“The user must be able to confirm successful verification of the prisoner on the mobile device screen and be able to immediately proceed to identify the next prisoner in a fully automated multi-transactional session and preferably without a user initiated manual trigger event,” with each ID taking less than one second. “The mobile application matched result shall then be able to retrieve locally stored identity, profile and prisoner information on the matched prisoner,” taken from the MPAC app, the solicitation states.
“Due to security requirements and operating safely within the COVID-19 environment, JPATS requires a touchless biometric solution,” as well, the document states.
One remaining question for the JPATS development team will be how many photos to include in the matching gallery.
While facial recognition is still maturing, different matching methods have different levels of reliability. Identification matching—or one-to-many—attempts to identify a person by matching a newly taken photo against a large, unassociated database, such as a criminal database of mugshots or the federal government’s Homeland Advanced Recognition Technology, or HART, database. This method has a higher failure rate, as the pool of photos tends to be much larger and there is no guarantee that the individual is included in the gallery.
The other major form of facial recognition is verification—or one-to-one—in which a novel photo is matched against a known photo of the person. This biometric technique is used in places like U.S. ports of entry, where the Customs and Border Protection agency uses in-motion cameras to capture travelers’ photos and match them against recorded travel documents, like a U.S. passport. This method tends to be highly accurate, especially when combined with high-quality cameras and lighting.
Other methods fall somewhere in between—one-to-a-small-many—such as the system developed by CBP and airlines to facilitate boarding. That system matches a new, in-motion photo with images connected to the flight manifest.
For JPATS, the decision has yet to be made whether to include the full roster of prisoners being transferred in a given day—7,000 on average—or “just the set of prisoners known to be on the current mission/trip that day,” the solicitation states.
Either way, “‘Matching thresholds’ must be configurable as well as provide [the] user with a result that can return multiple photo matching options when thresholds are not met. The matching thresholds must maintain an error rate lower than 15% during the custody exchange process.”
When fully deployed, the agency expects to install the app on about 1,000 devices, according to the solicitation.
Interested vendors must be able to answer the following eight prompts in detail:
- Required software, application, mobile application program interface (API) and/or web user interface(s) needed for functionality including details on any browser limitations or required service dependencies.
- Technical policy settings for the configuration specifications necessary for the function of the capabilities and to support the solution proposed.
- Centralized services, software or equipment needed for the mobile functionality including specification if infrastructure requirements exist.
- Connection requirements at the local, global and ecosystem levels.
- Power requirements, including impact to the mobile device.
- Artificial Intelligence code trustworthiness, performance and data interactions.
- Local device interactivity, inclusive of local device cache timeframe requirements.
- How the proposed solution meets federal regulatory and legal compliance with notation for any element of the proposed solution that cannot fully meet compliance.
The solicitation includes a host of other questions and requirements for vendors to consider.
“While a complete system that provides all of these capabilities is desirable, JPATS may consider solutions that can deliver partial capabilities or subsystem(s) that provide one or a few of the capabilities,” the solicitation states.
The contract will have a one-year base period through September 30, 2021, with four one-year license renewal options.
Bids are due by 3 p.m. September 25.
The latest solicitation follows another earlier this year in which the service requested proposals for biometric evasion experts to teach marshals the latest techniques in counter-biometrics and how to prevent suspects from avoiding or confusing the technology.