Gait Recognition Using Convolutional Neural Network with RGB-D Sensor Data

Abstract

Progress of an aging society is accelerated sharply in the world. Taking care of the aged is inevitable, but, especially in Japan, the population is shrinking and the people cannot support the aged enough. Here we consider watching over the aged by using technology. However, monitoring cameras are not accepted to many of the aged or their families. Thus we have been researching a robotic watching over system for the aged that do not give discomfort to them. Face recognition technologies are not suited for the watching over system, because of privacy infringing and of face recognition not working in the dark. As one of watching over functions, we have developed a gait recognition system using a convolutional neural network algorithm with RGB-D sensor data. The system does not use personally identifiable information like face, but still it can recognize the name of the aged even in the dark. We have developed a normalization method for RGB-D sensor data to make the subjects walk freely (not restricted to walk along a straight line, etc.). We have achieved accuracy of more than 91% for eight people classification.