Real-Time Slouch Detection and Human Posture Prediction from Pressure Mat

We present a system that is capable of real time slouch detection and posture prediction using a noninvasive pressure mat that lines the seat under a human user. Our contribution is that our final system requires neither wearable sensors nor external vision systems, although during system development, a motion capture system was used to collect ground truth data. We collected systematic seated posture data from four women aged 18 to 54 in order to build data-driven models using the seat pressure mat as sensory input. Due to the small number of subjects, it was found that individualized models were more effective than training one universal model for all participants. Our system is able to detect whether the user is slouching with up to 95% accuracy. Moreover, using just the pressure mat sensor readings as input, we successfully built models that could predict the positions of nine points on the body with an average position error of 8.84 cm. As long as one has access to a real-time stream of data from such a pressure mat, it is possible to perform real-time slouch detection and posture prediction with the pre-trained models investigated here.