Photo/Illutration An image used for an automobile model recognition experiment where AI detects cars in images shot by security cameras (Provided by the National Police Agency)

Whether the criminal activity you're dealing in is money laundering, stalking or selling drugs on social media, AI is coming for you, as Japan's police partner up more with the increasingly efficient multitasking technology.

The National Police Agency is exploring the potential and limits of artificial intelligence in investigations and security operations via trial runs. Some AI-based techniques have already yielded significant benefits in crime detection tests.

“We are aggressively forging ahead with trials for more advanced, improved police activity,” a senior NPA official said. “However, so that police officers can make final judgments on their own, we have no plans to leave everything to AI.”

The NPA in fiscal 2019 began a verification test of an AI system that analyzed images of dozens of models of vehicles.

Based on its analysis of still images of cars taken from security cameras near crime scenes, the system showed the five most likely car models in order of probability. 

The number of car models that the system is able to analyze is expected to rise to around 1,000 this fiscal year. The upgraded system to be tested will scan videos to identify target vehicles and extract scenes showing the same models.


An AI system used to aid money laundering investigations compares potentially illegal transactions reported from financial institutions and other organizations with cases that have previously been connected to crimes.

Examining a range of factors, the technology gauges the extent of risk to pinpoint which instances should be investigated first.

After a test of the system in fiscal 2019 verified the method’s effectiveness, the NPA is now looking to put it into full use by the end of this fiscal year.

The NPA conducted a similar experiment to detect suspicious behavior in crowds at large events for security purposes but encountered a few problems with it that still need to be ironed out.

People’s movements and other elements were analyzed based on videos taken in the experiment of places around the venues to pick out potential terrorist activity.

The findings revealed the precision level was higher for relatively uncrowded areas, but the suspicious behavior detection accuracy fell for sites where there were so many people that they were recorded overlapping each other in the videos.

An NPA official said that while the police plan to use the system as is, they hope to upgrade it.

“We will take advantage of the technology, such as combining it with a new technique to be developed from now on, keeping the outcomes accumulated through experiments in mind,” the official said.

From fiscal 2020, the NPA began a test to identify potentially dangerous behavior caught on video by security cameras at police boxes and other police facilities.

If the system detects suspicious acts such as people loitering, lingering in the same places or pulling out a weapon on the video, it sends that information out to local police headquarters and police stations.

The positive results of a trial of the system at 10 police substations and a few smaller counterparts for which Chiba and Gifu prefectural police are responsible led the NPA to utilize the technology to build a new system to detect and report suspicious acts and alert police officers.


In another verification test, complaints about stalkers and violence against partners were screened to determine the level of risk the person filing the complaint faced. The analysis was carried out based on a check list detailing the victims’ allegations and their relationships with the accused.

The findings suggested the system would help police officers make decisions on cases, and the NPA plans to review the points in its check list based on them.

An additional program is under way in fiscal 2021 that helps officers patrolling the streets easily refer to citizens’ names, drivers’ license information and their automobile registration numbers during questioning.

Specialized centers equipped with AI at local police headquarters record audio data of officers' requests for information.

The NPA is next looking into whether ways can be found for AI to filter out background noise and eliminate other potential bad telecommunications conditions that might cause officers' audio to be difficult to discern.


And that's not all. 

The NPA also expects to start using AI to sift through social media posts to catch communications linked to illegal drug trafficking in fiscal 2021. Information on buying and selling cannabis and other substances will be recovered and analyzed with attention paid to internet slang.

The system is anticipated to make it much easier for police to identify problematic posts and request their deletion.

Ko Shikata, a criminology professor at Chuo University who is knowledgeable about police operations, noted AI will help police to do more than just improve the efficiency of daily tasks.

“Use of AI will also make investigations and other activities fairer, because misjudgments can be prevented with it,” Shikata said.

Shikata also dispelled concerns about police relying on AI.

“The current way of using the technology simply aims to support officers’ decision-making, not to replace police officers,” he said. “As long as the final judgments are made by humans based on law, adoption of AI won't pose any adverse effects.”

The professor also noted that “combining knowledge on AI accumulated by police and the private sector may lead to the development of more advanced technologies needed by both sides.”