Research And Implementation Of An Action Recognition System Based On The Ubisense Positioning Platform

Action recognition has become the key research area of both human-computer interaction and ubiquitous computing, Traditional action recognition researches were mainly focused on computer vision, and have achieved a number of important research results. In recent decades, due to the rapid development of micro-machining technologies, the sensor is more and more small and smart, As such, it is becoming attractive to recognize activities by using sensors.This article employs the Ubisense positioning system to obtain location information of different parts of a pedestrian. Then a BP network based action recognition model and associated algorithms are thoroughly studied, on this basis, a real-time action recognition system is designed and implemented.Firstly, by placing UWB tags on the wrist, waist and ankle of a tester, the spacial position information of the three parts can be continuously collected during any activities performed by the tester. After data preprocessing, data segmentation, feature extraction, model training process and other processes, various classification models are evaluated, and as a result, the BP neural network is selected as the optimal one. The experimental results show that the overall recognition accuracy of the model for the six actions is above 88%.Secondly, the training of the proposed BP neural network model is further embedded into a web-based system based on the SSH framework and WEKA toolkit, and the system is able to provide online real-time activity recognition services.In summary, this paper explores a new action recognition method based on Ubisense sensors by employing location information of different parts of a human body. Feasibility and practicality of this method has been validated by experimental results and the real recognition system.