IARPA launches procurement process for video data to advance biometric tracking and re-identification
The Intelligence Advanced Research Projects Activity (IARPA) has issued a request for information (RFI) for data collection from camera networks to support computer vision research, particularly for biometric tracking and re-identification.
Significant advances in tracking and re-identification have been made over the past five years, IARPA says, but further advances to deal with the types of video data generated by real-world video networks used by public safety and law enforcement agencies require more extensive datasets for training and performance evaluation. To build those datasets, IARPA is looking for data from networks made up of 20 or more video feeds with differing positions, resolutions, and frame rates, with both overlaps and gaps in their fields of view. It is also seeking data from large areas, crowded urban or semi-urban environments, as well as specific data involving volunteers performing according to instructions and data that meets privacy and legal standards for public release.
The RFI asks seven questions, six of which are optional and deal with the partnerships, processes, and annotation experience of the submitting organization. The mandatory seventh question makes clear that an organization providing data collection may not be eligible to participate in future IARPA research involving that data.
Improved analysis capabilities for multi-camera video network footage could help with post-event crime scene reconstruction, protecting critical infrastructure and transportation facilities, military security, and special event operations, according to the RFI document.
IARPA and NIST also recently launched a $50,000 prize challenge for activity detection in extended video.