Based in the UK, our client specializes in offering comprehensive insurance solutions to vehicle owners across multiple regions, including England, Scotland, Wales, and Northern Ireland. They provide insurance plans covering everything from basic liability coverage to extensive full coverage for various types of vehicles (cars, motorcycles, vans, and trucks).
The client approached us to detect and label various types of vehicle damage (such as dents, rust, and scratches) in over 3000 images submitted for insurance claims. These annotated images were required to improve the accuracy and efficiency of their AI-based damage detection algorithms. The goal was to reduce the time taken for claims assessment & processing and enhance the image analysis capabilities of the defect detection model.
The images provided by the client were captured in different resolutions and lighting conditions. Accurately annotating such images presented a few challenges for our team, such as:
We employed a team of 10 experienced annotators specializing in working with the client's specified image annotation tool - CVAT. To overcome the labeling challenges and correctly annotate the provided dataset, we followed a strategic approach.
To ensure that annotators could distinguish dents from reflections caused by shadows, we created a comprehensive guide detailing all possible types of vehicle damage. To reduce false positives, we provided our team with two-day training and also implemented a cross-verification system where multiple annotators reviewed uncertain cases.
We established an extended workflow for annotating low-light images, including techniques like HDR image processing and histogram equalization for adjusting brightness and contrast to enhance damage visibility.
Our experts utilized digital zoom and image enhancement tools within CVAT to magnify areas of potential subtle damage and labeled the defects using the polygon annotation technique. We also developed a color-coding system for different damage types to enhance visual distinction during the annotation process.
To maintain consistency and accuracy across all annotations, we implemented a multi-level QA process, including intra-annotator and inter-annotator checks. Subject matter experts were involved throughout the annotation process, from defining labels for automated image labeling by CVAT to verifying the accuracy of the outcomes generated by the tool and enhancing the reliability of labeled datasets.
Throughout the project, we communicated constantly with the client, fostering a collaborative approach. We incorporated the client's feedback iteratively to promptly address any discrepancies, ensuring the annotated dataset met their stringent quality standards.
Before
After
Before
After
40% Improvement in the Accuracy of Damage Detection Algorithms The AI model trained on our labeled image dataset was able to identify diverse vehicle damages more accurately and efficiently.
30% Faster Claim Assessments Better AI model efficiency streamlined the claims assessment process, allowing the client to process claims faster, resulting in improved customer satisfaction.
Enhanced Risk Assessment and Fraud Prevention By improving the capability of the defect detection system to identify subtle vehicle damages, we helped the client significantly reduce fraud and improve risk assessment accuracy.
At SunTec India, we excel in managing complex annotation tasks with unparalleled precision. Our human-in-the-loop methodology guarantees high-quality labeled datasets for machine-learning models. Get in touch with us to understand how our image annotation services can significantly benefit your business.