This AI claims to predict crimes before they happen based on real-time CCTV analysis
Serving tech enthusiasts for over 25 years.
TechSpot means tech analysis and advice you can trust.
Calculating risks: Although the concept can be traced back to earlier literary works and philosophical discussions, the movie Minority Report brought the idea of authorities stopping crimes before they happen into the mainstream. Now, that vision may soon become a reality, thanks to researchers in South Korea. The country’s Electronics and Telecommunications Research Institute has unveiled “Dejaview” – an AI system that analyzes CCTV footage to detect and potentially prevent criminal activity.
Dejaview uses machine learning to analyze patterns and identify signs of impending crimes. It considers factors like time of day, location, past incident records, and other variables to assess the risk of something suspicious occurring.
According to a report by TechXplore, the core technology operates in two key ways. First, there’s a time/space-based prediction model that evaluates elements such as whether a crime previously occurred in a remote area late at night.
For instance, if a quiet, isolated location shares similar environmental factors with a past late-night crime, the system assesses a high risk of another incident.
Authorities can then proactively monitor those high-risk zones more closely through CCTV feeds to prevent incidents before they start and position response teams appropriately. In field tests working with local Seocho city data, this “predictive crime mapping” system demonstrated an accuracy of 82.8%.
The second component of Dejaview is called ‘individual-centered recidivism prediction.’ It zeros in on individuals considered “high risk” for repeating the same offenses. By tracking their movement patterns, the technology can analyze whether their behavior signals they might commit another crime soon.
As for how Dejaview acquired its smarts, the technology was trained on a massive dataset of over 32,000 CCTV clips capturing various incidents across a three-year span. The AI learned to recognize patterns from this data, and now applies that ‘knowledge’ to live scenarios.
Of course, the Orwellian implications of AI-powered crime prediction will surely stir debate, especially when it comes to tracking individuals. For now, ETRI appears to be limiting Dejaview’s application to public safety infrastructure like airports, energy facilities, factories, and national event monitoring. Commercial use for specialized security agencies is expected by the end of 2025.
South Korea isn’t alone in exploring this technology. Argentina has also established a new AI unit aimed at preventing, detecting, investigating, and prosecuting criminals using specialized algorithms. Argentina’s approach goes a step further by analyzing data beyond CCTV, including social media, websites, and even the dark web.
