Remote Identification and Tracking of Non-Cooperative Subjects (REMIT-NCS)

The ability to reliably track and recognize individuals inside a security perimeter is a critical component of next-generation distributed security systems. While state-of-the-art surveillance systems can reliably track pedestrians in sparse, static environments with minor occlusions and few moving subjects, performance degrades rapidly as scene complexity and crowd density increase. The problem becomes even more difficult when tracking individuals over long time scales, or across cameras with non-overlapping fields of view, a scenario which is unavoidable in most urban environments. Existing systems are also unable to re-acquire individuals who have been previously tracked in a separate location, but for whom recent track data is unavailable. To address these issues, we propose a system for Remote Identification and Tracking of Non-Cooperative Subjects (REMIT-NCS). REMIT-NCS extracts stable, descriptive signatures from tracked individuals in surveillance imagery by reconstructing a tracked individual's anthropometric parameters (shape and pose) and using these to produce higher-order, viewpoint-insensitive signatures from the individual's intrinsic attributes (e.g., physical build and motion characteristics) and extrinsic attributes (i.e., outward appearance). The combined intrinsic and extrinsic signatures are then compared to a database of similarly processed tracks to identify the features most suitable for supporting long-term tracking and re-acquisition. Persons moving from one camera view to another are then re-acquired by the system via the resulting discriminative models, enabling persistent tracking of individuals throughout the facility.