Classifying Elevators and Escalators in 3D Pedestrian Indoor Navigation Using Foot-Mounted Sensors

For quick and targeted rescuing of individuals in emergency situations indoors an accurate knowledge of the current floor-level by the injured and rescue personnel is essential. In this paper, we investigate detection and characterization of transportation platforms like elevators and escalators using foot-mounted inertial measurement units including magnetometer and barometer data for applications in 3D pedestrian navigation. Several data sets including elevator and escalator rides in various environments and allowing for superposed activities by the pedestrian are recorded for this purpose. A variety of features and the selection of feature subsets are analyzed for classifying among static environments, elevator environments (up/down), and escalator environments (up, down). The features are compared via an information gain metric and selected classifier concepts are analyzed with focus on fast response times necessary for dead-reckoning navigation. We identified features exploiting statistical characteristics of the magnetic intensity and the acceleration to be most promising taking into account fast response times. The resulting feature space is highly non-linear and is best approximated locally.