IARPA posts Camera Network Research Data Collection RFI
On March 26, the Intelligence Advanced Research Projects Activity posted a request for information for Camera Network Research Data Collection. Responses are due by 12:00 p.m. Eastern on May 10.
This RFI seeks capability statements relating to the collection of research data from multi-camera video networks in support of computer vision research. Over the past five years, there have been notable advances in computer vision approaches to facilitate tracking and re-identification of persons in security camera networks. However, the primary datasets available to the research community for algorithm training and performance evaluations, while incredibly valuable, are somewhat limited in subject count, camera network scope, and environmental factors, resulting in a disconnect between the data being leveraged by researchers and the types of video data that would exist in actual video networks utilized by public safety and law enforcement entities. Further research in the area of computer vision within multi-camera video networks may support post-event crime scene reconstruction, protection of critical infrastructure and transportation facilities, military force protection, and in the operations of National Special Security Events.
This RFI seeks approaches, capabilities, and previous experience related to the ability to carry out a potential video research data collection with the following characteristics:
· Data from a large camera network containing a minimum of twenty (20) video feeds of varying positions, views, resolutions, and frame rates with both overlapping and non-overlapping fields of view;
· Data captured over a large region (~10,000 sq. meters) with multiple intersections, paths for pedestrian foot traffic, and buildings entrances/exits;
· Data collected in an urban or semi-urban environment with distractors and occlusions representative of real-world conditions, such as signs, vehicles (parked and moving), trees, and other obstructions;
· Data collected over multiple days with varying illumination and weather conditions;
· Data involving a minimum of 5,000 pedestrians;
· Data involving a minimum of 200 subject volunteers injected into the camera network view and given instructions on how to behave and/or where to go in the camera network;
· Annotations for all video footage detailing ground-truth bounding boxes of each person in each frame of each camera view to facilitate research, development, test, and evaluation of person tracking and re-identification algorithms;
· Minimum video footage for the entire dataset totaling an estimated 960 hours (20 cameras x 12 hrs/day x 4 days);
· A dataset that is permissible to be released to the general research community under a privacy, legal, and policy approved data release process.
Full information is available here.