An Estimation of Pedestrian Action on Footbridges Using Computer Vision Approaches

Abstract

Vibration serviceability of footbridges is important in terms of fitness for purpose. Human-induced dynamic loading is the primary excitation of footbridges and has been researched with traditional sensors, such as inertial sensors and force plates. Along with the development of computer hardware and algorithms, e.g., machine learning, especially deep learning, computer vision technology improves rapidly and has potential application to the problem. High precision pedestrian detection can be realized with various computer vision methods, corresponding to different situations or demands. In this paper, two widely recognized computer vision approaches are used for detecting body center of mass and ankle movement, to explore the potential of these methods on human-induced vibration research. Consumer-grade cameras are used without artificial markers, to take videos for further processing and wearable inertial sensors were used to validate and evaluate the computer vision measurements.