The U.S. Retail Industry experiences over $32B in losses per year (2008 statistics) due to retail theft and frauds. Organized crime accounts for a staggering portion of these losses. Current video surveillance networks rely heavily on an eyes-on approach to alerting officials of specific human-related behaviors. Unfortunately, humans are unable to maintain focus and attention across several video monitors for an extended period of time. Improved video behavioral analytics will serve to alert loss professionals and security personnel to behavioral abnormalities, direct their attention to the monitor and area of interest, and allow them to engage with the shoplifter before departure from the store. This will have a dramatic impact on Industry and our overall Economic Security. Improved technologies in the "Commerce" environment are directly applicable to improvements needed to notify police and security forces of potential threats. Hence, VISITORS receives a lot of attention from several communities.
The VISITORS project goes a step beyond any video analytic work done to date, in the sense of promoting the advancement of video analytics that are able to predict that a criminal event is about to take place through the observation of specific behaviors. This is done by predictive analysis, using software to analyze a video stream and to identify suspicious behaviors that enable notification of security personnel while tracking the person of interest up to, and including, the shoplifting event. The VISITORS project will require the development of a distributable video data corpus of normal and abnormal activities encountered in a retail environment.
As a direct result of the June 2010 Technical Symposium several senior-level working groups were developed and began meeting in the latter portion of 2010. The active working groups are:
- Suspicious Behaviors which encompasses law enforcement, psychologists, senior loss prevention officials, computer vision experts, and scientists. Intense brainstorming produced a list of salient external, internal and group behaviors which may be precursors to a shoplifting event.
- Strategic Planning which is comprised of government only personnel and is looking at steps forward in evaluation protocols and planning.
- Video Data Collection which met to discuss the various issues involved in collecting and disseminating video for training and testing algorithms. Options for using existing retail video databases and alternatively creating scenarios, based on desired suspicious behaviors, are being pursued. The group discusses a myriad number of issues, not least being dealing with legal issues concerning use of Human Subjects in research.
The heaviest push is focused on receiving clearances for data access and dissemination. Human subject protection and personal identifying information concerns must be resolved. Much effort remains in developing scenario based data collection protocols.
Efforts to develop the VISITORS Evaluation plan and metrics for an initial pilot evaluation are underway.
NIST has brought together major stakeholders from the retail and security industries, computer vision technologists/developers, the research community, law enforcement, and government agencies in a common mission to advance the state-of-the-art in predictive video analytics. The focus of the VISITORS project is to advance technologies and methodologies used to detect persons engaged in suspicious activities as applied in the retail domain. The initial "meeting of the minds" was held via a technical symposium in June 2010 to identify interest and key issues among the stakeholders. The resulting roadmap indicating the pathway ahead is now available.