The Stack Archive

Hitachi develops AI for real-time people detection and tracking

Mon 27 Mar 2017

CCTV airport

Researchers at Japanese multinational Hitachi are developing a new image analysis system using artificial intelligence (AI) to detect and track individuals in real-time.

The team explained in a news release that the detection software can pick out an individual in real-time by categorising and searching a combination of over 100 external appearance and movement features, such as sex, clothing colour, or carried items.

The system is able to track individuals through analysis of the entire body image and extracts a series of images of that person. Using high-speed vector processing, an image of the same person can be extracted from several tens of thousands of recorded images in less than a second.

The researchers suggest that the technology could help detect and track a suspicious individual or a lost child in public areas such as cities or large facilities like airports and stations.

Screen Shot 2017-03-27 at 12.00.19Alongside eye-witness accounts, the system would be able to detect people which fit the description from within wide-area images taken by public security and surveillance cameras.

According to Hitachi, by analysing the entire image of the person detected, the system is able to follow that person across camera images even when only the rear-view is captured and the face cannot be seen, or when the person is captured from a distance.

‘In order to prevent emergencies…it is necessary to be able to immediately close-in on images of a suspicious person or a lost child based on eye-witness accounts from persons on the scene, and locate in real-time the whereabouts of the person sought after from the wide-area surveillance camera network,’ explained the research team.

They noted that the advanced AI functionality enables the system to distinguish an individual using various features such as sex, age, and clothing, as well as movements such as walking, running and bending. While previous systems would have had to calculate these items separately to detect the characteristics, the new method is able to simultaneously calculate multiple features – reducing calculation time to 1/40th compared with conventional models.


AI Asia crime Japan news research
Send us a correction about this article Send us a news tip