Human Interaction Recognition Based on Joint Sequence

摘要

Human interaction recognition based on skeleton data has attracted widespread attention due to its fast speed and robustness. Aiming at the current problem that the skeleton data is imaged and combined with the convolutional neural network for recognition, which cannot effectively model the video time-series relationship. An interaction recognition method for joint sequence images is proposed. First calculate the joint-joint distance features of a single frame, and then quantize them into a grayscale image every three frames. Then each grayscale image is sent to the convolutional neural network to extract the deep features, and finally send these features to the Long Short-Term Memory network for time series modeling to achieve the human interaction recognition. Experiments on the internationally published SBU Kinect interaction database have achieved a recognition rate of 96%, which verifies the effectiveness of the proposed algorithm