An indoor human action recognition method based on spatial location information

In indoor environments, identifying human actions is of great importance for various context-aware applications, such as smart home, smart healthcare, habitat monitoring, and so on. As a result, abundant methods and systems have been developed to recognize human actions by using different types of information, e.g., static images, surveillance videos, signals of inertial sensors, and etc. Different from existing works, this paper deals with the problem by making use of spatial location information of three different parts of a human body, which are derived via three UWB-RFID tags and a Ubisense UWB positioining system, and further implements a classification system based on a backpropagation (BP) neural network model to predict six ordinary human actions (i.e., stand, walk, run, lay down, squat, and jump). This model is trained based on a practical experiment. An experimental analysis based on the method of 5-fold cross validation reveals that the classification accuracy is nearly 80%, indicating that the proposed system is efficient.