A research group at the Institute of Scientific and Industrial Research, Osaka University has developed a gait authentication method that can be analyzed with a single walking image.
Gait authentication is a technology that authenticates an individual based on the individuality of the way of walking (step length, arm swing size, spine stretch, etc.), and unlike other biometric information such as fingerprints and faces, it is far away from the camera. As the only biometric information that can be used in places, it is being increasingly used for forensic science using security camera images.However, in the conventional method, the walking image of one cycle (about 1 second of 2 steps left and right) is mainly used, which causes a time lag until authentication and the whole body image is observed only for a moment in a crowded environment. There was a problem that it was difficult to use in the situation where it was not done.In addition, although a gait authentication method that inputs an arbitrary number of walking images has been proposed, there is a problem that the authentication accuracy is significantly reduced with one walking image.
Therefore, this group has developed a deep learning model that estimates the walking posture from one walking image and restores the walking image for one cycle.In order to combine this with various conventional gait authentication methods, by learning model parameters in combination with a deep learning model of gait authentication, one cycle is restored from one walking image and gait authentication is performed. Created a new technology.
When the authentication accuracy of this method was evaluated using a public walking video database of about 10,000 people, the recognition error rate was about 15/1 compared to the conventional method, and individuals when the number of registered people was about 5,000. The first-ranked identification rate was improved by about 1 times.
With this result, gait authentication is possible even in situations where only one walking image can be used or where real-time performance is required, and gait appraisal from a single walking snapshot and full-body image can be instantly displayed. It is expected to be applied to crime prevention and surveillance in a crowded environment where only can be seen.
Paper information:[Institute of Scientific and Industrial Research, Osaka University] "Future Science Investigation" Gait Appraisal
