Pedestrian Travel Time Estimation in Crowded Scenes

Shuai Yi, Hongsheng Li, Xiaogang WangICCV 2015 · OpenReview

Abstract

In this paper, we target on the problem of estimating the statistic of pedestrian travel time within a period from an entrance to a destination in a crowded scene. Such estimation is based on the global distributions of crowd densities and velocities instead of complete trajectories of pedestrians, which cannot be obtained in crowded scenes. The proposed method is motivated by our statistical investigation into the correlations between travel time and global properties of crowded scenes. Active regions are created for each source-destination pair to model the probable walking regions over the corresponding source-destination traffic flow. Two sets of scene features are specially designed for modeling moving and stationary persons inside the active regions and their influences on pedestrian travel time. The estimation of pedestrian travel time provides valuable information for both crowd scene understanding and pedestrian behavior analysis, but was not sufficiently studied in literature. The effectiveness of the proposed pedestrian travel time estimation model is demonstrated through several surveillance applications, including dynamic scene monitoring, localization of regions blocking traffics, and detection of abnormal pedestrian behaviors.

Introduction

In video surveillance and urban planning, knowing how long it takes a person to get from Point A to Point B is a vital metric for measuring traffic efficiency. However, in crowded environments like train stations or plazas, estimating this is a significant challenge for computer vision. Traditional methods rely on pedestrian tracking or re-identification, which often fail in dense crowds due to constant occlusions and fragmented trajectories.

This research argues that perfect individual tracking is not necessary to understand crowd behavior. Instead, the authors propose looking at the "global" state of the scene—specifically the density and velocity of the crowd—to predict statistical travel time.

Methodology

The proposed framework moves away from individual-level analysis toward scene-level regression.

Overview of the proposed framework: (a) S-D trajectories, (b) travel time distribution, (c) moving pedestrian features, and (d) stationary pedestrian features.
  1. Active Regions: For every source-destination (S-D) pair, the system creates a weighted "active region" map. This map highlights areas where pedestrians are most likely to walk, allowing the model to ignore irrelevant activity.
  2. Feature Engineering: Two specialized feature sets are designed:
  3. Moving Pedestrian Features: These capture how other walkers interfere with a path, considering location importance and walking direction.
  4. Stationary Pedestrian Features: These model the "blocking effect" of people standing still, accounting for group size and density.
  5. Regression Model: These features are fed into a second-order polynomial regression pipeline to map the current scene state to an estimated average travel time for a specific time window.

Key Insights

Statistical analysis revealed that travel time is strongly correlated with activities along the walking route but weakly correlated with the total number of people in the entire scene. This justifies the use of "Active Regions" to filter out noise.

Modeling active regions and stationary blocking groups (red) for a specific source-destination pair.

The paper highlights that moving and stationary people affect travel time differently. Moving pedestrians cause others to adjust their speed to avoid collisions, while stationary groups act as physical barriers that force pedestrians to take longer, tortuous detours. Because the method uses "tracklets" (short, fragmented bits of motion) rather than complete trajectories, it remains functional in extremely crowded scenes.

Results and Applications

The model was evaluated on two challenging real-world datasets: New York Grand Central Station (over 12,000 annotated paths) and the Shanghai Expo.

  • Accuracy: On the Grand Central dataset, the proposed method achieved an average error of 2.32 seconds, significantly outperforming the Social Force Model (7.14s) and standard pedestrian tracking (9.45s).
  • Dynamic Scene Monitoring: The model accurately tracks "dynamic surges" in travel time caused by temporary blockages, useful for real-time security alerts.
  • Abnormal Behavior Detection: By comparing actual travel time to the predicted "normal" time, the system can automatically flag individuals walking too slowly, running, or taking suspicious routes.

Conclusion

This approach provides a robust solution for travel time estimation in dense crowds where traditional tracking fails. By focusing on global scene features and active regions, it offers valuable insights for crowd management and behavior analysis. Future work may explore applications in crowd disaster prediction and public space layout optimization.

Summary updated

Abstract

In this paper, we target on the problem of estimating the statistic of pedestrian travel time within a period from an entrance to a destination in a crowded scene. Such estimation is based on the global distributions of crowd densities and velocities instead of complete trajectories of pedestrians, which cannot be obtained in crowded scenes. The proposed method is motivated by our statistical investigation into the correlations between travel time and global properties of crowded scenes. Active regions are created for each source-destination pair to model the probable walking regions over the corresponding source-destination traffic flow. Two sets of scene features are specially designed for modeling moving and stationary persons inside the active regions and their influences on pedestrian travel time. The estimation of pedestrian travel time provides valuable information for both crowd scene understanding and pedestrian behavior analysis, but was not sufficiently studied in literature. The effectiveness of the proposed pedestrian travel time estimation model is demonstrated through several surveillance applications, including dynamic scene monitoring, localization of regions blocking traffics, and detection of abnormal pedestrian behaviors. Many more valuable applications based on our method are to be explored in the future.
BibTeX
@inproceedings{yi2015pedestrian,
  author = {Shuai Yi and Hongsheng Li and Xiaogang Wang},
  title = {{Pedestrian Travel Time Estimation in Crowded Scenes}},
  booktitle = {IEEE/CVF International Conference on Computer Vision},
  year = {2015},
  url = {https://openreview.net/forum?id=rJ-Ab-MdWr},
}
Figure