Vehicles Detection System at Different Weather Conditions

Abdullah, Mustafa Najm; Ali, Yousra Hussein · 2021 · Crossref

DOI: 10.24996/ijs.2021.62.6.30

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 robust vehicle detection in Intelligent Transportation Systems (ITS) under varying weather and lighting conditions. The authors identify that existing systems often struggle with accuracy during complex weather events such as rain, fog, snow, and cloudiness, which degrade video visibility. To overcome these limitations, the study proposes an adaptive vehicle detection system that dynamically adjusts its preprocessing steps based on the detected weather condition, aiming to improve detection accuracy and reliability for traffic monitoring applications. The proposed methodology employs a multi-stage pipeline. First, the system classifies the weather condition of the input video as either "normal" or "complex" using a Random Forest Classifier (RFC). This classification relies on texture features extracted from the first two frames via the Gray Level Co-occurrence Matrix (GLCM), specifically calculating contrast, correlation, homogeneity, angular second moment, energy, and dissimilarity. Once the weather state is determined, background subtraction is performed using the Mixture of Gaussian 2 (MOG2) algorithm to isolate moving objects. The system then applies adaptive preprocessing: for normal weather, it uses Gaussian blur, morphological operations (dilation and erosion), and thresholding; for complex weather, it adds histogram equalization to enhance contrast and improve clarity. Finally, regions of interest are classified as vehicles or non-vehicles using a Support Vector Machine (SVM) classifier, also utilizing GLCM features. The system was evaluated using datasets comprising images for weather classification (122 normal, 81 complex) and vehicle classification (12,111 vehicle, 4,111 non-vehicle images from Stanford Cars and GTI datasets). Testing involved 42 videos for weather classification and eight realistic videos covering diverse conditions, including cloudy, foggy, rainy, snowy, and sunset scenarios. The weather classifier achieved 99% training accuracy and 95.23% testing accuracy. The vehicle detection system demonstrated high performance across conditions, with detection rates ranging from 87.7% in snowy night conditions to 100% in foggy night conditions. The overall average detection rate was 96.4%. Comparative analysis against previous methods showed the proposed system outperformed or matched existing approaches in most tested scenarios, particularly excelling in foggy and heavy traffic conditions. The significance of this work lies in its adaptive approach to preprocessing, specifically the integration of histogram equalization for complex weather, which effectively mitigates visibility issues. The study concludes that combining GLCM-based feature extraction with adaptive preprocessing and dual-classifier models (RFC for weather, SVM for vehicles) provides a robust solution for ITS. The results confirm the system's ability to maintain high accuracy under challenging environmental conditions, suggesting potential for future expansion into vehicle tracking and counting for comprehensive traffic management.

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.

• weather rain fog snow