A real-time system for vehicle detection with shadow removal and vehicle classification based on vehicle features at urban roads

Atouf, Issan; Al Okaishi, Wahban Yahya; Zaaran, Abdelmoghit; Slimani, Ibtissam; Benrabh, Mohamed · 2020 · Crossref

DOI: 10.11591/ijpeds.v11.i4.pp2091-2098

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 traffic monitoring in urban environments, where complex traffic flows and slow vehicle movements complicate data collection. Traditional electronic sensors, such as loop detectors, are limited by small detection zones and invasive installation requirements. To overcome these limitations, the authors propose a real-time vision-based system for vehicle detection, tracking, and classification at urban intersections. The system aims to estimate traffic density and optimize traffic light timing by accurately counting vehicles across three categories: cars, bikes, and trucks. The methodology consists of three sequential phases. First, vehicle detection is performed using background subtraction via the Mixture of Gaussians (MoGs) algorithm, which separates moving objects from the static background. A critical innovation in this phase is a custom shadow removal algorithm designed to eliminate shadows that distort vehicle dimensions and cause misclassification. This algorithm utilizes Canny edge detection on both the foreground mask and the grayscale source image, followed by an XOR operation and background edge subtraction to isolate true vehicle edges. Second, vehicles are tracked through consecutive frames using a feature-based method that correlates centroids based on minimum Euclidean distance. Third, classification occurs when vehicles cross a designated line, utilizing geometric features—specifically height, width, and aspect ratio—to categorize them. The system includes logic to handle horizontal and vertical occlusions by splitting detected regions when dimensions exceed specific thresholds. Experimental validation was conducted using video data from stationary cameras in Casablanca, Morocco, comprising two scenes with resolutions of 240x320. The system demonstrated high accuracy, achieving an average counting accuracy of 96.78% across both scenes. Specifically, Scene 1 yielded accuracies of 97.45% for cars, 96.2% for bikes, and 94.73% for trucks, while Scene 2 achieved 97.59%, 94.7%, and 100% respectively. A comparative study against other surveillance systems indicated that the proposed method’s average accuracy of 96.78% outperforms several existing approaches, particularly those relying on optical flow or frame differencing. The significance of this work lies in its robust, real-time capability to classify multiple vehicle types in crowded urban settings without requiring complex machine learning models. By effectively removing shadows and handling occlusions through geometric analysis, the system provides reliable data for intelligent transportation systems. The results confirm that the proposed pipeline is efficient and accurate, offering a viable solution for alleviating traffic congestion through automated density estimation and adaptive traffic light control.

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.

• vehicle conspicuity