Sensor fusion and human behavioral pattern analysis for vehicle occupancy detection

Context sensitivity is a key feature in modern vehicles, which includes seat occupancy detection. Current systems are prone to misdetection or are sometimes non-existent as for rear passengers. Our approach is to look at human behavior while getting into, sitting inside and getting out of a vehicle. Occurring sensor patterns, including door opening and closing, damper and seat movement and seat belt use are analyzed and fused. These sequences are then modelled by a Dynamic Bayesian Network. Over 360 cases including more than 50 subjects are recorded and used for training. A 10-times cross validation is performed which gives an overall accuracy of above 90%.