Developing a framework for promoting physical activity in a Boccia game scenario

Abstract

The traditional keyboard has been replaced by tactile screens or other types of implicit interfaces. An example of such interface is the Microsoft Kinect system, which makes use of a depth sensing system to enable the user-machine interaction. Allied to the physical activity, this hardware may be used to track and monitor the user when he/she is exercising, but also to diminish a sedentary lifestyle. In this paper, it is proposed a system based on such non-wearable and wearable devices to monitor the elderly while playing Boccia. This system allows to recognise the game movements as well as it registers other physiological variables of the player. The results show the comparison of different methods and approaches to recognise two main gestures used during a Boccia game. Firstly, the non-wearable and wearable approaches were compared by training an SVM model using data from Kinect and the Pandlet. From the results obtained, the accuracy for the model with Kinect was higher. Then, in order to improve the gesture recognition, several models were trained with the accelerometer data from the Pandlet. The results showed that the RBF SVM had better results achieving a cross-validation accuracy of 95%.