An insurer struggled to manually verify vehicle damage for claims. Learn how Aya’s computer vision model facilitated rapid detection and classification.
Overview
Our client, a leading insurance company in Africa, wanted to use advanced technology to streamline the process of verifying insurance claims. They aimed to create a computer vision (CV) model capable of detecting and identifying vehicle damage through images taken with smartphones. This model needed to handle various types of vehicles, including cars, vans, motorbikes, and buses.
Challenge
The task was to gather and label 85,000 images of damaged vehicles from West Africa. The client had detailed requirements for how these images should be collected and categorised by the type of damage and the vehicle involved. After collecting the images, they needed to be quickly annotated and tested for quality before being delivered.
Solution
Aya Data’s experienced teams in Ghana and Sierra Leone collected the required images. Once collected, our specialists in Ghana carefully annotated the images using precise polygons to highlight damaged areas.
To speed up the process, we developed and trained a custom model to pre-label the images. This pre-labeling step significantly accelerated the overall labelling process for the entire dataset.
Results
Thanks to Aya Data’s efforts, the client developed a highly accurate computer vision model for detecting and classifying vehicle damage. This allowed their insurance teams to rapidly verify if a claim matched the vehicle type and damage described.
Moreover, our pre-labeling approach enabled the client to launch their model three weeks earlier than planned. The client is now looking forward to working with us again on future projects to enhance their models for various scenarios.
Ready to improve your insurance claims process? Contact us today to discover how our precise image annotation can boost efficiency and accuracy in claim validation.
Disclaimer: At Aya Data, we prioritise the confidentiality of our clients and their projects. As a result, specific names and identifiable details are kept anonymous in our case studies.
