Traffic Monitoring System for Vehicle Detection in Day and Night Conditions

Ibtissam, Slimani; Abdelmoghit, Zaarane; Issam, Atouf · 2023 · DOAJ

DOI: 10.2478/ttj-2023-0020

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 reliable vehicle detection in Intelligent Transportation Systems (ITS) under varying lighting conditions. While existing computer vision solutions perform well during the day, they often fail at night due to weak illumination. The authors propose a unified, real-time traffic monitoring system capable of detecting vehicles in both daytime and nighttime environments using a static roadside camera. The system aims to improve road safety and traffic control by providing high-accuracy detection without relying on expensive sensors like lasers or radars. The proposed method employs a two-stage pipeline: candidate generation and candidate verification. For candidate generation, the system applies a Two-Dimensional Discrete Wavelet Transform (2D-DWT) using Haar wavelets to denoise and compress input images, retaining low-frequency sub-images. Background subtraction via a Gaussian Mixture Model is then used to isolate foreground objects, with periodic background updates to handle lighting and weather changes. During the day, connected components are extracted directly as vehicle candidates. At night, the system filters for bright white (headlights) and red (taillights) components, grouping them into pairs based on color, size, horizontal alignment, and proximity to form potential vehicle candidates. In the verification stage, a pre-trained Convolutional Neural Network (CNN), specifically the Inception v3 model adapted via transfer learning, classifies these candidates as vehicles or non-vehicles. Experiments were conducted using five video sequences (three daytime, two nighttime) on a dual-core ARM Cortex-A9 processor with 1 GB of memory, utilizing C++, OpenCV, and TensorFlow. The system achieved high precision and recall rates, outperforming four comparative methods from the literature. For daytime detection, the proposed method reached precision rates between 97.96% and 98.83% and recall rates between 99.59% and 100%. For nighttime detection, it achieved precision rates of 98.24% and 98.78%, and recall rates of 98.24% and 98.49%. Furthermore, the system demonstrated real-time performance, processing approximately 23 frames per second for both day and night modes. The significance of this work lies in its ability to provide a cost-effective, efficient, and accurate solution for continuous traffic surveillance. By integrating traditional image processing techniques (2D-DWT, background subtraction) with deep learning (CNN), the system overcomes the limitations of single-mode detectors. The results confirm that the approach is robust against environmental changes and capable of real-time operation on embedded hardware, making it suitable for practical deployment in intelligent transportation applications.

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