By The Japan News
The technology is able to identify individuals based on surveillance camera images that captures them walking, even if only for a moment. It is expected to be applied in criminal investigations.
The method is called gait recognition. According to Yagi, who specializes in visual information processing, gait recognition has been used in criminal investigations in Japan since 2009 and has been introduced as evidence at trials. For accurate results, it was necessary to have at least four steps’ worth of walking data, including two steps each on the left and right foot.
The team used data from videos of about 5,000 people walking, which was collected for experimental purposes, and analyzed it with deep learning — machine learning algorithms that detect patterns from large amounts of data. It succeeded in extrapolating four steps of walking data from a single gait image by finding such characteristics as the way a person steps forward and their posture. The team said it could identify individuals with 80% accuracy based on this extrapolation.
NEC Corp. Fellow Hitoshi Imaoka, who specializes in biometrics, said: “The idea of reproducing data of walking is quite interesting. Chances to obtain a sufficient amount of gait images are usually limited. The technology can be said to be effective in identifying individuals based on little information.”