Human posture classification is an important tasks in medical applications, i.e., patient monitoring, ulcer prevention, and conduct diagnostic. We propose a system for posture recognition of lying-down human bodies using a low-resolution pressure sensor array. A support vector-machine was used to perform the classification of pressure maps. Three databases were constructed in order to represent the pressure maps: pressure raw-data, HOG and SIFT image descriptor vectors. It was found that the image descriptors have improved complexity time to build the classification models rather than using raw pressure maps. Experimental results was performed in order to control a robotic hospital bed.

In this paper we show a methodology for bodies classification in lying state using HOG descriptor and pressures sensors positioned in a matrix form (14 x 32 sensors) on the surface where bodies lie down. it will be done in real time. Due to current technology a limited number of sensors is used, wich results in low resolution data array, that will be used as image of 14 x 32 pixels. Our work considers the problem of human posture classification with few information (sensors), applying digital process to expand the original data of the sensors and so get more significant data for the classification, however, this is done with low-cost algorithms to ensure the real-time execution.

No abstract