IDOD-YOLOV7: Image-Dehazing YOLOV7 for Object Detection in Low-Light Foggy Traffic Environments

Yongsheng Qiu, Yuanyao Lu, Yuantao Wang, Haiyang JiangSensors 2023 · DOI

TL;DR

This paper introduces IDOD-YOLOV7, a novel framework for object detection in low-light foggy traffic environments. It tackles the degradation of detection performance in such conditions by jointly optimizing an image defogging module (IDOD, combining AOD and SAIP) with the YOLOV7 object detector, leading to improved accuracy and robustness.

What Problem Does This Paper Address?

Autonomous driving systems rely heavily on accurate perception, with object detection being a critical component. However, adverse weather conditions like low-light fog severely degrade the performance of standard object detection algorithms. Existing methods often struggle to effectively balance image enhancement (to recover visibility) and object detection, leading to missed detections and reduced accuracy. This paper aims to bridge this gap by proposing a unified approach that enhances image quality and performs object detection simultaneously in challenging low-light foggy scenarios.

Methods

The core of the proposed solution is the IDOD-YOLOV7 framework, which integrates three key components:

  • AOD (All-in-One Network for Dehazing and Beyond) Module: This module performs initial image defogging by optimizing parameters within an atmospheric scattering model. It uses parallel convolutional layers to compensate for information loss during processing.
  • SAIP (Self-Adaptive Image Enhancement) Module: This module further enhances the defogged image. Unlike traditional methods that manually tune parameters, SAIP uses a lightweight CNN (CNN-PP) to predict adaptive hyperparameters (white balance, gamma correction, contrast, and sharpening) for each image. This allows for more precise image enhancement tailored to the specific low-light fog conditions.
  • YOLOV7 Detection Module: The state-of-the-art YOLOV7 object detector is used for the final object detection task.

The entire framework is trained end-to-end, allowing the defogging and enhancement modules to work synergistically with the detector.

Datasets

  • FTOD (Foggy Traffic Object Detection) Dataset: The authors created this dataset by physically fogging scenes and using artificial synthesis. It aims to address the domain transfer problem often encountered when training on synthetic data and deploying on real-world foggy images. FTOD contains 5100 pairs of fog-free and fogged images across various fog concentrations.
  • MS COCO Dataset: Used for pre-training the YOLOV7 model before fine-tuning on the FTOD dataset.
  • SOTS and HAZE Datasets: Used for evaluating the performance of the image dehazing module.

Key Insights and Ideas

  • Joint Optimization is Crucial: The paper strongly advocates for the simultaneous learning of image defogging, enhancement, and object detection. This integrated approach is shown to be more effective than sequential processing.
  • Adaptive Enhancement Matters: Manually tuning image enhancement parameters is suboptimal. Using a CNN to predict these parameters adaptively based on image characteristics (brightness, color, fog concentration) significantly improves results.
  • Real-World Data is Key: The creation of the FTOD dataset highlights the importance of real-world data for robust performance, especially in overcoming domain shift issues from synthetic datasets.
  • Low-Light Fog is a Unique Challenge: The paper emphasizes that standard object detection models perform poorly in low-light foggy conditions, necessitating specialized solutions.

Key Results

  • Improved Detection Accuracy: IDOD-YOLOV7 achieved superior mean Average Precision (mAP) compared to various state-of-the-art object detection algorithms (Fast R-CNN, Faster R-CNN, SSD, YOLOV3, YOLOV7) on low-light foggy traffic datasets. For instance, on the "Heavy" fog test set, IDOD-YOLOV7 reached an mAP of 0.6036, outperforming YOLOV7 (0.5387).
  • Enhanced Image Quality: The AOD+SAIP module demonstrated strong performance in objective metrics like PSNR and SSIM, outperforming other defogging methods on the FTOD dataset. Qualitative results showed improved contrast, saturation, and reduced artifacts.
  • Reduced Missed Detections: Ablation studies showed that the combined enhancement and defogging approach (IDOD-YOLOV7) significantly reduced missed detections compared to using only enhancement or only defogging.
  • Efficiency: While introducing additional modules increases computational load (4% time complexity increase compared to YOLOV7), the detection speed still saw a notable increase of 10.7%.

Limitations and Future Work

  • Domain Transfer: While the FTOD dataset helps, the paper acknowledges that the performance on real-world foggy images can still be affected by domain transfer issues.
  • Fog Depth and Concentration Estimation: The authors suggest that further improvements could be achieved by explicitly incorporating fog depth and concentration estimation into the object detection framework. This would allow for even more tailored processing.

Summary updated

Abstract

Convolutional neural network (CNN)-based autonomous driving object detection algorithms have excellent detection results on conventional datasets, but the detector performance can be severely degraded in low-light foggy weather environments. Existing methods have difficulty in achieving a balance between low-light image enhancement and object detection. To alleviate this problem, this paper proposes a foggy traffic environment object detection framework, IDOD-YOLOV7. This network is based on joint optimal learning of image defogging module IDOD (AOD + SAIP) and YOLOV7 detection modules. Specifically, for low-light foggy images, we propose to improve the image quality by joint optimization of image defogging (AOD) and image enhancement (SAIP), where the parameters of the SAIP module are predicted by a miniature CNN network and the AOD module performs image defogging by optimizing the atmospheric scattering model. The experimental results show that the IDOD module not only improves the image defogging quality for low-light fog images but also achieves better results in objective evaluation indexes such as PSNR and SSIM. The IDOD and YOLOV7 learn jointly in an end-to-end manner so that object detection can be performed while image enhancement is executed in a weakly supervised manner. Finally, a low-light fogged traffic image dataset (FTOD) was built by physical fogging in order to solve the domain transfer problem. The training of IDOD-YOLOV7 network by a real dataset (FTOD) improves the robustness of the model. We performed various experiments to visually and quantitatively compare our method with several state-of-the-art methods to demonstrate its superiority over the others. The IDOD-YOLOV7 algorithm not only suppresses the artifacts of low-light fog images and improves the visual effect of images but also improves the perception of autonomous driving in low-light foggy environments.
BibTeX
@inproceedings{qiu2023idod,
  author = {Yongsheng Qiu and Yuanyao Lu and Yuantao Wang and Haiyang Jiang},
  title = {{IDOD-YOLOV7: Image-Dehazing YOLOV7 for Object Detection in Low-Light Foggy Traffic Environments}},
  booktitle = {Sensors},
  year = {2023},
  doi = {10.3390/s23031347},
}
Figure