TSBOW

TSBOW Logo

T S B O W

Traffic Surveillance Benchmark for Occluded Vehicles under Various Weather Conditions

AAAI 2026

[Slides] [Poster] [Presentation]

Ngoc Doan-Minh Huynh, Duong Nguyen-Ngoc Tran, Long Hoang Pham, Tai Huu-Phuong Tran,
Hyung-Joon Jeon, Huy-Hung Nguyen, Duong Khac Vu, Hyung-Min Jeon, Son Hong Phan,
Quoc Pham-Nam Ho, Chi Dai Tran, Trinh Le Ba Khanh, Jae Wook Jeon

Automation Lab, Sungkyunkwan University, South Korea
Explore TSBOW Paper Paper arXiv Supplementary Huggingface Hugging Face Dataset GitHub GitHub

Overview

198

Processed Videos

32h

Duration

3.2M

Total Frames

71.1M

Semi-Annotated Instances

48K

Manual-Annotated Frames

1.1M

Manual-Annotated Instances

Scenario: Road and Intersection
Scenario: Special Cases and Disaster

Scenes

Scenes from the TSBOW dataset, comprising 198 videos recorded across four distinct scenarios spanning all seasons (sunny/cloudy, haze/fog, rain, snow) over a year. The dataset emphasizes adverse weather conditions and densely populated urban areas with heavy traffic, addressing significant challenges in image degradation and vehicle occlusion.

Intersections under different viewpoints in diverse weather conditions

Subset for Datasets comparison
scenes compare
Selected scenes for comparison with other datasets

To ensure a fair comparison, we created a subset of medium-scale scenes distinct from the TSBOW dataset, featuring unique road structures and vehicle characteristics. While snowy conditions were recorded in Suwon, additional videos capturing normal, haze, and rain conditions were collected in Seoul. Unlike UA-DETRAC, which includes only fine- and medium-scale videos captured by a color camera, and UAVDT, which focuses on medium- and coarse-scale drone footage, TSBOW encompasses fine, medium, and coarse scales. Therefore, the comparison subset comprises medium-scale scenes, included in the UAVDT, UA-DETRAC, and TSBOW datasets.

TSBOW All Scenes

Because of bandwidth limitation, only images are shown in this website. The demo videos are provided in TSBOW Scenes.

Annotation Pipelines

Detailed overview of the data collection and annotation pipeline. The process commences with the recording and categorization of videos during the data collection phase. Subsequently, the videos are preprocessed and allocated to a team of annotators for manual labeling. Next, a state-of-the-art model is fine-tuned to automatically annotate the remaining frames. The resulting annotations are then verified against predefined labeling criteria. Finally, the annotated instances are aggregated and undergo post-processing to finalize the dataset.

The detailed description and examples for each class are provided in Class_Definition.pdf. Other documents related to the TSBOW dataset can be found at Documents folder.

Object Categories
vehicle Categories
Visualization of annotated instances of different classes

vehicle Categories
Instance statistics

Statistics

Suwon Recording Locations

Attribute Distribution

Occlusion Distribution (#annotations)

Traffic Distribution (#videos)

Challenges

Challenges by Disaster scenarios

Car accident due to snow weather

Examples of Road Types and Scales

Challenges by Weather Conditions

Download

Choose the download option that suits your needs

The TSBOW dataset is distributed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. By completing the Terms and Conditions form, users acknowledge and agree that the dataset will be used solely for research purposes.

Submission Guidelines
  1. TSBOW - Terms and Conditions Form
    • Download and fill out the TSBOW - Terms and Conditions form.
    • Ensure all User Information fields are completed.
    • Provide a description of your intended use of the dataset.
    • Check the agreement box and insert your handwritten signature.
    • Save as PDF file. Renaming as: {TSBOW}_{Application-Date}_{Huggingface-Username}.pdf (i.e. TSBOW_20260120_ngochdm.pdf)
  2. Requirement Email
    • The subject format: [TSBOW Access Requirement] {Your Name} - {Affiliation}
    • The email body includes your HuggingFace account information (username and email). We will verify this information against the access requirements on Hugging Face before approval.
      • Send email to all our addresses: jwjeon@skku.edu, ngochdm@skku.edu, automation.skku@gmail.com
  3. Send Request on HuggingFace
    • Press "Agree and send request to access TSBOW" button on HuggingFace. Our team may take 2-3 days to process your request.

TSBOW Full Dataset

Complete dataset with 198 videos, extracted frames and all manually/semi-labeled annotations

Hugging Face Hugging Face GitHub GitHub

Experiments

Model performances under different weather conditions

Validation Models
model performances
Model performances after training 100 epochs and validating with imgsz=1280 on manually labeled test set.
Comparison of Traffic Surveillance Datasets
Comparison of traffic surveillance datasets
Model performances when training on different datasets
Comparison of traffic surveillance datasets
Comparison of Traffic Surveillance Datasets
Comparison of car class across traffic surveillance datasets
Models performance for car across different metrics on the comparison set

Ablation Studies

Object Classes
YOLOv12x performance across different classes
YOLOv12x performance across different classes.
Data Characteristics
Influence of dataset characteristics on object detection performance
Influence of dataset characteristics on object detection performance.

Abstract and Future Works

Abstract Global warming has intensified the frequency and severity of extreme weather events, which degrade CCTV signal and video quality while disrupting traffic flow, thereby increasing traffic accident rates. Existing datasets, often limited to light haze, rain, and snow, fail to capture extreme weather conditions. To address this gap, this study introduces the Traffic Surveillance Benchmark for Occluded vehicles under various Weather conditions (TSBOW), a comprehensive dataset designed to enhance occluded vehicle detection across diverse annual weather scenarios. Comprising over 32 hours of real-world traffic data from densely populated urban areas, TSBOW includes more than 48,000 manually annotated and 3.2 million semi-labeled frames; bounding boxes spanning eight traffic participant classes from large vehicles to micromobility devices and pedestrians. We establish an object detection benchmark for TSBOW, highlighting challenges posed by occlusions and adverse weather. With its varied road types, scales, and viewpoints, TSBOW serves as a critical resource for advancing Intelligent Transportation Systems. Our findings underscore the potential of CCTV-based traffic monitoring, paving the way for new research and applications. The TSBOW dataset is publicly available at the following link.
Code -- https://github.com/SKKUAutoLab/TSBOW.
Conclusion and Future Works This study introduces the Traffic Surveillance Benchmark for Occluded vehicles under various Weather conditions (TSBOW), a comprehensive, semi-automatically annotated traffic surveillance dataset designed to improve monitoring system training, particularly under extreme weather conditions such as heavy haze and snow. Collected across all seasons and diverse road scenarios, TSBOW comprises 32 hours of footage from 198 videos, encompassing a variety of road types and scales, and providing multiple viewing angles for vehicles and pedestrians. The dataset includes over 3.2 million frames, each annotated with weather conditions and scenarios, alongside detailed object annotations derived from extracted images. Capturing complex, high-density scenes of vehicles and pedestrians in crowded urban settings, TSBOW features approximately 71.1 million bounding boxes across eight distinct traffic participant classes. As a robust resource for traffic surveillance research, TSBOW offers substantial potential to deepen insights into traffic dynamics and support advancements in intelligent transportation systems. The initial version focuses on daytime traffic flow under varying weather conditions. Future versions will include ground truth annotations for nighttime scenarios and additional computer vision tasks, such as multi-object tracking, semantic segmentation, vehicle counting, and speed estimation, to further enhance its utility.

Cite Our Work

If our research is helpful to you, please cite our paper
using the following BibTeX format

BibTeX Icon 
@article{Huynh2026TSBOW, 
    title={TSBOW: Traffic Surveillance Benchmark for Occluded Vehicles Under Various Weather Conditions}, 
    author={Huynh, Ngoc Doan-Minh and Tran, Duong Nguyen-Ngoc and Pham, Long Hoang and Tran, Tai Huu-Phuong and Jeon, Hyung-Joon and Nguyen, Huy-Hung and Khac Vu, Duong and Jeon, Hyung-Min and Phan, Son Hong and Pham-Nam Ho, Quoc and Tran, Chi Dai and Khanh, Trinh Le Ba and Jeon, Jae Wook}, 
    journal={Proceedings of the AAAI Conference on Artificial Intelligence}, 
    volume={40}, 
    number={7}, 
    url={https://ojs.aaai.org/index.php/AAAI/article/view/37439}, 
    DOI={10.1609/aaai.v40i7.37439}, 
    year={2026}, 
    month={Mar.}, 
    pages={5239-5247} 
}

Acknowledgement

This work was supported by the Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. 2021-0-01364, An intelligent system for 24/7 real-time traffic surveillance on edge devices).