Related Experiment Video
Updated: Jun 17, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
499
Automated vehicle damage classification using the three-quarter view car damage dataset and deep learning approaches.
Donggeun Lee1, Juyeob Lee2, Eunil Park2,3
1Department of Artificial Intelligence, Sungkyunkwan University, Seoul 03063, Korea.
Heliyon
|August 6, 2024
Summary
We introduce the Three-Quarter View Car Damage Dataset (TQVCD) to address challenges in automated vehicle damage classification. This dataset simplifies labeling and enhances data accessibility for improved classification accuracy.
Area of Science:
- Computer Vision
- Machine Learning
- Automotive Engineering
Background:
- Automated vehicle damage classification is crucial for industries managing large vehicle fleets.
- Existing challenges include a lack of public datasets and complex data construction processes.
- Three-quarter vehicle views offer rich information for damage assessment.
Purpose of the Study:
- To introduce a novel dataset, the Three-Quarter View Car Damage Dataset (TQVCD), for vehicle damage classification.
- To provide a dataset emphasizing simple labeling, data accessibility, and comprehensive damage information.
- To evaluate deep learning model performance on this new dataset and validate its industrial relevance.
Main Methods:
- Development of the Three-Quarter View Car Damage Dataset (TQVCD) with distinct classes for orientation and damage type.
- Evaluation of five pre-trained deep learning architectures (ResNet-50, DenseNet-160, EfficientNet-B0, MobileNet-V2, ViT) using binary classification models.
- Implementation of a model ensemble method to improve classification robustness.
- Expert interviews with used-car platform professionals to assess industrial applicability.
Main Results:
- The TQVCD dataset effectively captures diverse vehicle perspectives and damage types.
- Deep learning models demonstrated varying performance, with ensemble methods enhancing robustness.
- Expert validation confirmed the necessity and utility of the TQVCD dataset for industrial applications.
- The dataset facilitates efficient data collection and damage classification, reducing manual labeling efforts.
Conclusions:
- The TQVCD dataset is a valuable resource for advancing automated vehicle damage classification research and applications.
- The dataset's design addresses key limitations of existing resources, promoting wider adoption.
- The study validates the practical utility of automated damage classification in the used-car industry.
