Using computer vision to extract pedestrian behaviour indicators

November 20, 2025

Understanding pedestrian behaviour in urban environments is crucial for effective planning. However, traditional data collection methods are labour-intensive and limited in scope, often focusing on movement-related activities. This thesis explores the use of computer vision to extract micro-scale pedestrian behaviour indicators from static street videos, addressing both the challenges of data collection and lack of tools for analysing non-movement activities such as resting, lingering, and socializing. A modular framework is developed that integrates a computer vision model for object detection (YOLO), a tracking algorithm (DeepSORT), and clustering algorithm (DBSCAN) to identify groups of pedestrians. Three variants of the YOLOv12 family of models were tested: the pretrained medium and extra large models, as well as a medium model trained on a custom dataset. From this analysis, indicators are calculated, such as pedestrian flows, speeds, trajectories and group sizes, heatmaps of stay spots, lingering durations and ratios of moving vs stationary pedestrians. A homography step to calibrate pixel distances to real-world measurements is also incorporated. The framework is applied to a case study at Instituto Superior Técnico's Alameda campus, where 196 videos recording the parking lot in front of the central building are analysed. Results show the trained model outperforms pretrained versions in precision and recall. The indicators reveal that the study area is used as a crossing zone with limited lingering and low group prevalence with peak flows around 12:00 and 17:00. The framework demonstrates strong potential for scalable, automated analysis of pedestrian behaviour.

Authors in the community:

• 

  F

  Francisco Maças

  ist199760

Supervisors of this institution:

• 

  G

  Gabriel Costa Valença

  ist194281

• 

  F

  Filipe Manuel Mercier Vilaça e Moura

  ist14188

civil-engineering - Civil engineering

por - Portuguese

Instituto Superior Técnico