Time Series Human Motion Prediction Using RGB Camera and OpenPose

抄録

The report projects that by 2050, the population aged 60 and above will reach 2.1 billion. This aging society is more likely to suffer from locomotive syndrome. In order to reduce the spread of locomotive syndrome, it is best to increase awareness before the citizens become elderly. We propose the system to predict human motion as the first step to realize the locomotive syndrome estimation. Previous researches were using the Kinect camera which has a depth sensor that the camera used to detect the pose of a human body. However, in this research we are using an RGB camera as a reliable alternative. We set a goal to predict 1 second ahead of the motion which includes simple motions such as hand gesture and walking movement. We used OpenPose to extract the features of a human body pose including 14 points. YOLOv3 is used to crop the main feature in the frames before OpenPose process the frame. Distance and direction which are calculated from the features by comparing two consecutive frames as the input of Recurrent Neural Network Long Short-term Memory (RNN-LSTM) model and Kalman Filter. Mostly, Kalman Filter show better accuracy then RNN-LSTM and based on the human motions, motion such as hand gesture and moving to the right side are easier than more complex motion like hand gesture and moving to the left side. We confirmed the validity of RGB-camera based method in the simple human motion case from the result.