Practitioner Centric Video Analytics, Final Summary Overview

• NCJ Number: 250272
• Author(s): Ming-Ching Chang; Jixu Chen; Siwei Lyu; Peter Tu
• Date Published: October 2015
• Length: 17 pages
• Grant Number(s): 2013-IJ-CX-K010
• Agencies: NIJ-Sponsored
• Publication Type: Research (Applied/Empirical); Report (Study/Research); Report (Grant Sponsored); Grants and Funding

Annotation

Using practitioner feedback as input, this project sought to develop three new video analytics technologies: 3D video representation and event summarization front-end; one-shot learning for action recognition; and person-specific face recognition.

Abstract

Three-dimensional video representation and event summarization front-end allows users to observe view-independent 3D event summarizations with highlights and greater contextual support. One-shot learning for action recognition enables recognition of novel events by using a single user-provided example. One-shot learning for action recognition enables recognition of novel events using a single user-provided example. Person-specific face recognition makes it possible to search against a face catalog by specifying certain distinguishing facial features, such as visible scars and hair styles. The achievement of this research agenda and insights gained from interaction with the practitioner community provide a new form of "recognition at a glance," allowing for greater understanding of complex multi-camera imagery. With one-shot learning for action recognition, analytics engines will be able to detect new types of behaviors with a single example. The surveillance systems of the future will be able to keep pace with the evolving demands on law enforcement. Also, by considering person-specific physical cues, it will be possible to increase the detection of persons of interest. Scholarly products and in-process documents of this project are listed. 3 figures and appended supplementary project details