Detecting and Counting Vehicles from Satellite Imagery for Internal Displacement Monitoring in the Middle East

Abstract

Access to reliable mobility data is crucial for understanding human movement patterns, especially in conflict-affected regions where conventional data sources are often scarce or unreliable. Vehicle detection and counting in satellite imagery offer a promising proxy for tracking population movements, as shifts in vehicle distributions are tied to shifts in population distributions. This approach is particularly valuable in the Middle East, a region marked by high conflict activity and minimal cloud cover, allowing consistent satellite imaging. However, realizing this potential requires addressing key challenges, including partial or missing data, vehicle detection accuracy, and efficient spatial-temporal modeling. This thesis tackles these challenges via the development of two large-scale benchmark datasets. First, we introduce the Vehicles in the Middle East (VME) dataset, specifically curated for vehicle detection in high-resolution satellite images from the region. Sourced from Vantor, the VME dataset spans 54 cities across 12 Middle Eastern countries, with over 4,000 image tiles and 100,000 annotated vehicles, significantly improving detection accuracy in this region. Additionally, we present the Car Detection in Satellite Imagery (CDSI) dataset, the largest benchmark dataset of its kind, integrating images from multiple sources to enhance global vehicle detection, and our experiments demonstrate that models trained on CDSI achieve substantial improvements in car detection performance worldwide. Building on these advances, we conduct a large-scale case study on the Syrian civil war to assess the utility of vehicle detection for internal displacement monitoring. Using high-resolution satellite imagery covering 14 Syrian cities from 2009 to 2023, we apply our state-of-the-art region-adapted vehicle detection model and integrate it with a partial coverage correction method, enabling robust estimation of car counts from incomplete satellite observations. Using IDP reports from official sources, we evaluate the relationship between vehicle dynamics and displacement patterns across Syrian cities. Our validation results show that when official reports and satellite images overlap in time period and geography, there is a good directional agreement between changes in car counts and reported internal displacement. However, we also observe that these two sources are often complementary in geographic and temporal coverage, indicating that our remote sensing approach could help fill data gaps. These findings show that satellite-based vehicle detection can provide a valuable and scalable complement to traditional displacement monitoring methods in data-scarce conflict settings, as illustrated through the Syrian case.

Date of Award	2026
Original language	American English
Awarding Institution	HBKU College of Science and Engineering