A New Behavior Recognition Method of Nursing-Care Robots for Elderly People

Based on the optical flow technology, a convolutioal neural network (CNN) is proposed for nursing-care robots to perform the behavior recognition task, which considered both static and dynamic information during human motions, thus it is more accurate than the traditional CNN. Firstly, a behavior processing method, combining with Lucas-Kanade optical flow technology, is elaborately designed and tested. In this method, the limitation of static processing method existed in CNN is solved well, then the method is applied to a CNN model for behavior recognition task. Simulation experiment has been carried out, indicating that this method can achieve a higher recognition accuracy and obtain a good recognition effect successfully.