Road Traffic Monitoring from Aerial Images Using Template Matching and Invariant Features

Qureshi, Asifa Mehmood; Mudawi, Naif Al; Alonazi, Mohammed; Chelloug, Samia Allaoua; Park, Jeongmin · 2024 · Crossref

DOI: 10.32604/cmc.2024.043611

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of accurate road traffic monitoring using aerial images, specifically aiming to overcome the limitations of stationary cameras and the high computational costs associated with deep learning models. While Unmanned Aerial Vehicles (UAVs) offer flexible, large-area coverage, existing methods often struggle with complex road scenes, low frame rates, and varying altitudes. The authors propose a lightweight, computationally efficient algorithm that combines traditional image processing techniques for vehicle detection and tracking, avoiding the need for specialized hardware or extensive training data. The methodology involves a multi-stage pipeline applied to two publicly available datasets: the Traffic flow A1 Beekbergen Deventer dataset (Netherlands) and the Vehicle Aerial Imaging from Drone (VAID) dataset (Taiwan). First, images are converted to grayscale and georeferenced to embed coordinate information. To reduce computational load, a manual masking technique eliminates irrelevant background areas. Noise reduction is achieved through a combination of median filtering, Sobel and Canny edge detection, and the Hough line transform to remove road lane markings. Vehicle detection is performed on the first frame of every five-image burst using a blob detection algorithm, which identifies vehicles based on convexity, color, size, and inertia, utilizing dynamic thresholding to handle varying vehicle sizes. Tracking is executed on the subsequent four frames using template matching to locate potential matches, followed by Scale Invariant Feature Transform (SIFT) feature matching to refine results and minimize false positives by incorporating motion information. Experimental results demonstrate the effectiveness of this approach. On the A1 Motorway dataset, the system achieved an 87% accuracy rate for vehicle detection and an 80% accuracy rate for tracking. For the VAID dataset, which features diverse angles and illumination conditions, the model achieved 86% detection accuracy and 78% tracking accuracy. The system was implemented on a standard PC with an Intel Core i7 processor and 16 GB of RAM, highlighting its low memory and computational requirements compared to deep learning alternatives. The significance of this work lies in providing a transparent, lightweight solution for aerial traffic monitoring that does not rely on opaque deep learning models. By combining blob detection with template matching and SIFT features, the proposed method offers a flexible framework that can be easily modified for different traffic scenarios. The study confirms that traditional computer vision techniques, when properly combined, can achieve high accuracy in complex aerial environments while remaining accessible for deployment on standard hardware.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• visual search