Korean researchers develop AI CCTVs to detect, predict criminal activities

A Korean research team has succeeded in developing a technology that can detect and predict signs of criminal activities utilizing real-time CCTVs and AI technology. With this new technology, CCTV is now evolving beyond its initial role of surveillance, opening up its potential as an effective way of ‘preventing’ crimes.

Electronics and Telecommunications Research Institute (ETRI) has recently announced that they have completed the development of “Dejaview,” an innovative technology that uses CCTV footage, crime-related statistics, position determination technology, etc., to detect signals of potential crimes and predict the possibility of crime occurrence.

The researchers of “Dejaview” revealed that they have focused on the tendency of crimes to occur repeatedly based on a certain pattern, such as type, method, location, time, etc. “Dejaview” is an AI-based technology that measures and predicts the risk of potential crimes by comparing current environmental/social patterns with past criminal cases.

ETRI’s “Dejaview,” which predicts the occurrence of criminal activities through a “chance-based approach,” can be classified into two different technologies according to the application target and method. First, time/space-oriented crime prediction technology is based on the analysis of criminal activities and their chances depending on factors such as time and space.

For example, if a specific combination of environmental factors is observed in a remote/quiet location where a crime occurred in the past late at night, the system will view the risk of a crime taking place at that location as being very high. Through this, the system will prevent the occurrence of a crime in advance. The “Dejaview” system is expected to be mainly used by the police and integrated CCTV control centers of local governments nationwide.

By using a set of data related to past criminal records and crime-related statistics, the AI of Dejaview automatically analyzes real-time CCTV footage to compare and measure current situations and its similarities with past crime situations. In addition, the user can also obtain a better understanding of the current situation, such as immediately identifying and tracking suspected crimes/disasters (i.e., stalking, collapsing, early fires, drug trafficking, smuggling, etc.)

In collaboration with Seocho-gu, ETRI has developed a predictive crime map (PCM) by analyzing a total of 32,656 CCTV footage that includes incidents/criminal activities which took place in the region over the past three years from 2018. Based on past crime-related statistics, the system shows areas with high probability of criminal activities by categorizing the incidents by time, location, type of incident, etc.

Furthermore, PCM presents the real-time crime risk on the main screen. Using this feature, the ‘crime controller’ can get a grasp of the current level of existing crime risk by region/area. In other words, selective monitoring can become a viable option, putting more focus on specific areas with a high risk of criminal activities. According to the performance test conducted by the Telecommunications Technology Association (TTA), the crime prediction performance of the predictive crime map was measured at 82.8%

ETRI also developed an individual-centered recidivism prediction technology as the second key feature of Dejaview. This technology is strictly applied to high-risk individuals who are extremely likely to commit another crime. It analyzes the risk of individuals based on their movement pattern.

The current electronic supervisory system determines whether a specific individual has violated their restriction based on their positional information. If ETRI’s technology can be implemented to the existing system, it can periodically check whether that individual has violated their movement restrictions due to work-related reasons, etc., utilizing AI technology to conduct an in-depth analysis of their behavioral patterns and risk of recidivism.

ETRI’s research team plans to develop the source technology of Dejaview into a specialized system that focuses on safety-related services. They plan to expand it into various safety-related services, including field-specific crime prediction systems, proactive response systems for major infrastructure/key facilities (airports, energy facilities, factories, etc.), safety systems for major national events, preemptive response systems against dangerous acts/intentions for personal security purposes, etc.

To this end, it is necessary to develop a world-class video security AI technology that can accurately analyze, understand, and infer situations by collecting visual information from real-time CCTV footage/images.

Based on the developed Dejaview technology, the researchers of ETRI will work with local governments and agencies to provide crime signal detection and prediction services specific to various security sites and purposes. In addition, ETRI plans to expand the current technology into the development of an AI-based electronic monitoring technology that can detect and preemptively respond to the deviant behavior of high-risk individuals who are subject to electronic monitoring. The commercialization of Dejaview is expected to take place at the end of next year.

Kim Geon-woo, a senior researcher at ETRI’s AI Convergence Security Research Department, stated, “We have successfully developed a video security technology that uses CCTV information to predict and prevent the possibility of criminal activities, going beyond the simple detection of crime occurrence. Through this technology, we have now laid the technical foundation that will allow us to take the leap to the next level and the development of a futuristic, advanced social safety system.”