Proposal on Prediction of Pedestrian Falls Using Knowledge Graph Embedding

Abstract

This paper proposes a method to predict the risk of pedestrian falls by using knowledge graph embedding, assuming that data about actual pedestrian falls in a room or a public space will become available in the future owing to the spread of various sensing devices. In the proposed method, a pedestrian's situation is classified into three types: falling, passable, and impassable, and constructs a knowledge graph that represents each situation as a triple. Using the knowledge graph embedding obtained from the constructed knowledge graph, pedestrian's situation is estimated by link prediction. Experiments are conducted using data generated by a simple simulation to show the effectiveness of the proposed method. Experimental results show that prediction accuracy is improved by adding knowledge about pedestrians and roads.