It is said that the way people walk reveals their individuality. In fact, the walking posture of each person varies: some walk with a hunchback, some with a small gait, and some with a big wave of the hand.

Identity recognition of persons is a challenge in security cameras and surveillance cameras. This is because these cameras often have difficulty recognizing a subject's face, depending on the shooting environment and image quality settings.

In particular, in the infrared monitoring camera project that we are working hard on in this laboratory, while the use of an infrared array sensor instead of a general visible light camera allows for privacy, infrared light cannot capture personal characteristics such as facial features, so in order to identify the subject so that the care level and monitoring level can be set individually, a technology for determining the same person that can distinguish between silhouette images is required.

This project aims to perform person authentication based on the way a person walks by fitting the subject to a skeletal model using skeletal estimation based on deep learning technology based on video data captured by visible light cameras and infrared sensors, deriving feature points from the estimated positions of joint parts, and clustering the changes in the time series. We are conducting research with the aim of Since images of people taken by cameras and sensors are not always taken from the same direction, it is important to calculate feature values that do not fluctuate even when images are taken from various directions.

Pose estimation of walking experiment using treadmill
Identifying walking style by extracting feature-points