Tested in several stores in Japan, a surveillance camera system – combined with artificial intelligence that analyse customers’ body language – tracks shoplifters.

Our bodies speak volumes about our intentions. This is the reality behind AI Guardman, a surveillance camera system combined with artificial intelligence that analyzes body language to identify shoplifters. This system has been developed for more than two years by Japanese telecom giant NTT East and start-up Earth Eyes Corp, which presented it in a press release at the end of May.

A technology from the American Carnegie Mellon University

This tool uses open source technology developed by the American Carnegie Mellon University. Image analysis algorithms, driven with a database, scan live video streams and estimate the posture of visible bodies on images. The algorithms then seek to find in the attitudes captured those typical of shoplifters: such as looking around, looking for the location of cameras and spotting blind spots. If “suspicious behaviour is detected”, AI Guardman alerts merchants via a connected application.

With limitations, of course: according to The Verge, NTT East admitted that the device made several “common errors”, including that of considering as suspects undecided customers – those who take an item, put it back on the shelf and then take it again – and vendors who replenish the shelves. Moreover, the company has not published any statistics on the software’s error rate. This rate would be difficult to estimate anyway since the actual intent of the client may remain unknown. The seller can have a deterrent effect just by offering his help to the customer spotted by AI Guardman.

These limits do not stop NTT. According to The Verge, the Japanese group intends to market its cameras detecting thieves at the end of the month and targets 10,000 stores equipped within three years.