Biometric Identification Using Video of Body Silhouette Captured from Overhead

One of the most actively studied natural user identification methods would be the ones that use the gait. Although gait-based methods are quite promising, their applications often require rather large open space for implementation. We wanted to propose methods that are appropriate for home or small office. We used video data of a user captured by a camera installed right overhead of the user. We obtained image sequences while users were pushing a button and return to the initial position without moving the feet placement. From the image sequences, we obtained users’ silhouette and the bounding box of the silhouette that change in time. We defined 26 features from the silhouette itself and length and width of the bounding box, then selected six features through the analysis of principal components. We achieved recognition accuracy of 99% for 12 users using a nearest-neighbor classifier.