DualCam Traffic Light Dataset

A video compilation of training set ground truth

Overview

DualCam dataset is a benchmark traffic light dataset which covers urban and sub-urban areas. It consists of 2250 annotated images of 1920x1080 resolution with 8321 instances.

Download

The train set, test set 1, test set 2, test videos of DualCam traffic light dataset can be downloaded from following google drive links.

Samples - 15 image pairs (26 MB)
Train set - 1032 images (479 MB)
Test set 1 - 335 image pairs (547 MB)
Test set 2 - 478 image pairs (751 MB)
Test videos - 28 video pairs (3.63 GB)

Annotations

The annotations are given in PASCAL VOC XML and YOLO annotation format.

Statistics

Traffic light class	Train set	Test set	Total
Green	1198	1251	2449
Red	565	901	1466
Green-up	426	495	921
Empty-count-down	537	225	762
Count-down	346	396	742
Yellow	452	246	698
Empty	222	469	691
Green-right	115	171	286
Green-left	55	105	160
Red-yellow	66	80	146

Dataset folder structure

.
├── samples
│   ├── images
│   ├── xml
│   └── yolo
├── test_1
│   ├── images
│   ├── xml
│   └── yolo
├── test_2
│   ├── images
│   ├── xml
│   └── yolo
├── test_videos
└── train
    ├── images
    ├── xml
    └── yolo

Image Visualizers

We provide 2 visualizers to visualize images with the bounding boxes.

Install

Clone repo and install requirements.txt in a Python>=3.7.0 environment

git clone https://github.com/harinduravin/DualCam.git #clone
cd DualCam
pip install -r requirements.txt  # install

To visualize image use following scripts.

single-frame visualizer to visualize images with bounding boxes.

python3 single_image_visualizer.py

dual-frame visualizer to visualize image pairs in test set with bounding boxes.

python3 sync_image_visualizer.py

Camera details

Camera	Imaging sensor	Lens	FoV (Vertical)	FoV (Horizontal)	Frame rate (fps)
Narrow-angle camera	Basler daA1920-30uc (S-Mount)	Evetar Lens M13B0618W F1.8 f6mm 1/3” lens	34.5⁰	48⁰	30
Wide-angle camera	Basler daA1920-30uc (CS-Mount)	Theia SY125A/SY125M Lens	109⁰	125⁰	30

Please consider citing our paper, if you find our dataset valuable for your research

@inproceedings{dualcam2023,
  author={Jayarathne, Harindu and Samarakoon, Tharindu and Koralege, Hasara and Divisekara, Asitha and Rodrigo, Ranga and Jayasekara, Peshala},
  booktitle={2023 International Conference on Machine Learning and Applications (ICMLA)}, 
  title={DualCam: A Novel Benchmark Dataset for Fine-Grained Real-Time Traffic Light Detection}, 
  year={2023},
  pages={1778-1783},
  doi={10.1109/ICMLA58977.2023.00270}
}