Abstract
This study presents a gait recognition system using recurrent neural networks (RNNs) for human identification. Our system extracts the spatio-temporal features of distances between head and joints from joint positions obtained by Kinect sensor. The features are invariant under the distance between a walking subject and Kinect sensor. To confirm our system performance, we conducted tests with data of 10 persons. The data were divided into training and test data-sets. The RNNs were trained for classification using the training data set. Our proposed system achieved an average recognition accuracy rate of over 89% for the test data set.
