The spatial distribution of people in a building could provide valuable information for an effective guidance in emergency. This paper aims to develop a method to estimate the number of walking people in staircases using infrared human detection sensors. First, we constructed a walking behavior model and a sensor response model using the survey data observed in an emergency drill. Next, using these two models, we generated a large volume training data on human flows and corresponding sensor responses for a neural network model. Finally, we demonstrated the validation of the model through the comparison with the survey data.
