Car damage assessment system : a cloud-based serverless web application powered by deformable convolutional network plus | |
| Author | Sitthiwat Damrongpreechar |
| Call Number | AIT RSPR no.DSAI-25-01 |
| Subject(s) | Automobiles--Collision damage--Estimates--Data processing Deep learning (Machine learning) Cloud computing |
| Note | A research study submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Data Science and Artificial Intelligence |
| Publisher | Asian Institute of Technology |
| Abstract | Accurate and efficient vehicle damage assessment remains a critical challenge in the automotive insurance industry. Manual inspections are time-consuming, costly, and prone to error, contributing to claims leakage and delayed settlements. As insurance providers increasingly seek automation to improve service speed and reduce operational costs, there is growing interest in AI-powered assessment systems.This study addresses these challenges by developing a cloud-based, serverless web application that performs real-time car damage assessment using deep learning. A Deformable Convolutional Network Plus (DCN+) model was trained on the Car Damage Detection (CarDD) dataset, which provides comprehensive annotations across six damage categories to support accurate instance segmentation. The system was deployed using AWS Fargate and integrated with a React frontend, while GPT-4o was used to generate interpretable natural language reports.Thesystemachievedstrongtechnical anduser-centered performance. Thetrained model obtained a mean Average Precision (mAP) of 0.593 for bounding boxes and 0.559 for segmentation. All users in a usability study completed tasks successfully, with high ratings for clarity, efficiency, and ease of use. Replacing SageMaker with a containerized deployment on ECS improved cost control and compatibility but introduced cold-start delays and required more complex orchestration. User feedback also revealed opportunities to enhance the experience beyond core model accuracy—such as clearer post assessment navigation, more visible report access, and support for optional manual damage confirmation—highlighting the importance of transparency, control, and interface design in fostering trust.This research demonstrates the feasibility and effectiveness of combining deep learning and serverless cloud infrastructure for automated car damage assessment. It contributes a scalable, explainable, and user-centered solution for streamlining claims processing, particularly in emerging markets like Thailand where digital transformation in insurance remains limited. |
| Year | 2025 |
| Type | Research Study Project Report (RSPR) |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Data Science and Artificial Intelligence (DSAI) |
| Chairperson(s) | Chaklam Silpasuwanchai |
| Examination Committee(s) | Chantri Polprasert;Chutiporn Anutariya |
| Scholarship Donor(s) | Royal Thai Government Fellowship |
| Degree | Research Studies Project Report (M. Eng.) - Asian Institute of Technology, 2025 |