ABSTRACT
The vibration serviceability assessment of footbridges exposed to high density crowds is often imperative for the design. Due to the lack of representative operational loading data, current models for crowd-induced loading remain unvalidated. This lack of data can partly be ascribed to the fact that the existing techniques to directly register human-induced loading are infeasible to apply for a high-density crowd. Instead, this contribution is taking a first step towards the indirect measurement of crowd-induced loading by identifying the individual walking trajectories of a high-density crowd from video footage. A unique case study of a footbridge exposed to real pedestrian traffic with a density of 1 person/m2 is presented. Based on the collected video footage, the detection of each individual pedestrian in the crowd is obtained by employing an image segmentation algorithm. Next, each detection is associated to the corresponding track by using a motion-based prediction model. Subsequently, the obtained tracks are converted to world coordinates using two different setups: mono and stereo vision. The accuracy of the identified tracks is investigated. Finally, the flow characteristics are analyzed. The error propagation of measurement noise of the detection was observed to result in an accuracy in an order of magnitude of centimeters for the stereo-vision setup. Also, the error was found to be nearly constant for the entire bridge deck. In case of the single-view setup, the maximum error is higher than the stereo setup, especially for oblique viewing angles. A track obtained by mono-vision conversion has nearly the same accuracy as a track obtained by stereo vision provided that the true height of each pedestrian is accounted for. The identified mean walking velocity in the crowd is found to be in agreement with a widely-applied speed-density model in literature.
