Damage Detector

Industry: Insurance

Insurance Platform

Overview

Real-world damage detection from videos/photographs of vehicles being in an accident. Insurance companies have to deal with a lot many claims daily & have to employ surveyors to deal with estimations of cost. Which in the result is prone to human error & may result in some discrepancies. To avoid such cases, we were tasked to handle the damage detection part for 4 wheelers.

Technology

  • Deep Learning
  • Transfer Learning
  • Neural Networks
  • Computer Vision
  • Image Processing

Language

  • Python

Key Technical Challenges:

  • Working with all types of damages like glass breakage, doors slammed, dents, scratches to name a few.
  • Annotating the images & getting to a sufficient amount of data points for training
  • Creating clustered environment for training due to huge processing taken by image kind of data
  • Quick inference for production system.
  • Making a generalized model for all categories of 4 wheelers

Business + Technical Points:

  • Quick response from trained model
  • Scalable & adaptable model for heavy vehicles
  • Easy training pipeline & responsive dashboard.

Result:

  • We were able to identify & meet requirements with 92% accuracy.