Action recognition algorithm based on skeletal joint data and adaptive time pyramid

Human action recognition technology plays an crucial role in the fields of video surveillance, video retrieval, sports medicine and human-computer interaction. Slow research and application of this technology limited to complex environments and plasticity of human action. As a new sensor, Kinect provides a new idea for human action recognition, which can synchronously obtain data of skeleton joint points from target. In this paper, we propose a human action recognition method using skeletal joints data. The motion and static information of human action are firstly fused as feature and skeletal vector is used to construct motion model which can describe variation of human action after feature extraction. Then the model is introduced into adaptive time pyramid to capture global and local information; furthermore, skeletal joints feature in each period of time is processed. Finally, kernel extreme learning machine is used for human action recognition. Experimental results show that our work successfully achieves skeleton information in comparison with other methods.