Vision-Based Trajectory Reconstruction in Human Activities: Methodology and Application

Area of Science:

• Civil Engineering
• Biomechanics
• Computer Vision

Background:

• Civil engineering structures face increasing vibrations from human activities.
• Accurate crowd-induced load modeling requires understanding crowd dynamics and human-human interactions.

Purpose of the Study:

• To present a vision-based methodology for reconstructing individual movement trajectories.
• To capture crowd dynamics in both small groups and large-scale events for vibration analysis.

Main Methods:

• Integrates color-based segmentation for instrumented individuals and deep learning for uninstrumented crowds.
• Employs homography-based projection to map image coordinates to real-world space.
• Applies the framework to data from urban running events and controlled experiments with static and dynamic cameras.

Main Results:

• The framework reliably reconstructs individual trajectories across various conditions, including walking and running.
• Achieves scalable monitoring of human activities and provides high-resolution spatio-temporal data.
• Demonstrates applicability in both research and applied engineering contexts.

Conclusions:

• Vision-based methods offer practical, non-intrusive tools for analyzing human-induced loading.
• The developed methodology enhances the study of human-human interactions and crowd dynamics.
• Findings support improved load modeling for civil engineering structures susceptible to vibrations.