Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Low-Light Image Enhancement Framework for Improved Object Detection in Fisheye Lens Datasets

Dai Quoc Tran, Armstrong Aboah, Yuntae Jeon, Maged Shoman, Minsoo Park, Seunghee Park

TL;DR

The paper tackles object detection in fisheye urban surveillance under low-light conditions by introducing a Low-Light Image Enhancement Framework that combines transformer-based image enhancement (NAFNet) and night-to-day conversion (GSAD) with illumination-based data clustering. It trains an ensemble of detectors (Co-DETR, YOLOv8x, YOLOv9e) on enhanced day-like images and applies a robust post-processing pipeline using super-resolution with Dual Aggregation Transformer and Weighted Box Fusion to improve precision. Empirical results on FishEye8K show notable improvements over baselines, with Our-2 achieving $AP$ around $0.489$ on the validation set and the approach placing 5th in the 2024 AI City Challenge Track 4 with a F1 score of $0.5965$, demonstrating practical gains in accuracy and robustness for fisheye-based traffic monitoring. The work provides public code and highlights a viable path toward resilient, real-world intelligent traffic management using fisheye cameras under variable illumination.

Abstract

This study addresses the evolving challenges in urban traffic monitoring detection systems based on fisheye lens cameras by proposing a framework that improves the efficacy and accuracy of these systems. In the context of urban infrastructure and transportation management, advanced traffic monitoring systems have become critical for managing the complexities of urbanization and increasing vehicle density. Traditional monitoring methods, which rely on static cameras with narrow fields of view, are ineffective in dynamic urban environments, necessitating the installation of multiple cameras, which raises costs. Fisheye lenses, which were recently introduced, provide wide and omnidirectional coverage in a single frame, making them a transformative solution. However, issues such as distorted views and blurriness arise, preventing accurate object detection on these images. Motivated by these challenges, this study proposes a novel approach that combines a ransformer-based image enhancement framework and ensemble learning technique to address these challenges and improve traffic monitoring accuracy, making significant contributions to the future of intelligent traffic management systems. Our proposed methodological framework won 5th place in the 2024 AI City Challenge, Track 4, with an F1 score of 0.5965 on experimental validation data. The experimental results demonstrate the effectiveness, efficiency, and robustness of the proposed system. Our code is publicly available at https://github.com/daitranskku/AIC2024-TRACK4-TEAM15.

Low-Light Image Enhancement Framework for Improved Object Detection in Fisheye Lens Datasets

TL;DR

The paper tackles object detection in fisheye urban surveillance under low-light conditions by introducing a Low-Light Image Enhancement Framework that combines transformer-based image enhancement (NAFNet) and night-to-day conversion (GSAD) with illumination-based data clustering. It trains an ensemble of detectors (Co-DETR, YOLOv8x, YOLOv9e) on enhanced day-like images and applies a robust post-processing pipeline using super-resolution with Dual Aggregation Transformer and Weighted Box Fusion to improve precision. Empirical results on FishEye8K show notable improvements over baselines, with Our-2 achieving around on the validation set and the approach placing 5th in the 2024 AI City Challenge Track 4 with a F1 score of , demonstrating practical gains in accuracy and robustness for fisheye-based traffic monitoring. The work provides public code and highlights a viable path toward resilient, real-world intelligent traffic management using fisheye cameras under variable illumination.

Abstract

This study addresses the evolving challenges in urban traffic monitoring detection systems based on fisheye lens cameras by proposing a framework that improves the efficacy and accuracy of these systems. In the context of urban infrastructure and transportation management, advanced traffic monitoring systems have become critical for managing the complexities of urbanization and increasing vehicle density. Traditional monitoring methods, which rely on static cameras with narrow fields of view, are ineffective in dynamic urban environments, necessitating the installation of multiple cameras, which raises costs. Fisheye lenses, which were recently introduced, provide wide and omnidirectional coverage in a single frame, making them a transformative solution. However, issues such as distorted views and blurriness arise, preventing accurate object detection on these images. Motivated by these challenges, this study proposes a novel approach that combines a ransformer-based image enhancement framework and ensemble learning technique to address these challenges and improve traffic monitoring accuracy, making significant contributions to the future of intelligent traffic management systems. Our proposed methodological framework won 5th place in the 2024 AI City Challenge, Track 4, with an F1 score of 0.5965 on experimental validation data. The experimental results demonstrate the effectiveness, efficiency, and robustness of the proposed system. Our code is publicly available at https://github.com/daitranskku/AIC2024-TRACK4-TEAM15.
Paper Structure (16 sections, 2 equations, 9 figures, 3 tables, 1 algorithm)

This paper contains 16 sections, 2 equations, 9 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multiple/Two-stage detectors
  4. Single-stage detectors
  5. Transformer-based detectors
  6. Methodology
  7. Effective Data Pre-processing Strategy
  8. Detection Models for Training
  9. Robust Post-processing Strategy
  10. Experiments
  11. Dataset
  12. Evaluation Metrics
  13. Results and Discusion
  14. Comparative analysis on validation set
  15. Model performance on CVPR challenge set
  16. ...and 1 more sections

Figures (9)

  • Figure 1: The proposed approach has three stages. 1) effective data preprocessing strategy (NAFNET chen2022simple; GSAD hou2024global), 2) detection models for training ( Co-DETR zong2023detrs;YOLOv8aboah2023real;YOLOv9wang2024yolov9 ), and 3) robust post-processing strategy (DAT chen2023dual; WBF solovyev2021weighted)
  • Figure 2: Compare NAFNet chen2022simple image enhancement algorithm with different trained dataset.
  • Figure 3: Compare mean value from all scenario images.
  • Figure 4: Converting night time image to day-like image using GSAD model.
  • Figure 5: Using super-resolution to increase image size by a factor of four. a) raw image, b) super-resolution image
  • ...and 4 more figures