Cox Automotive

Recognizing the challenges arising from a lack of standardized training and clear quantifiable metrics, our client sought a solution to ensure uniform vehicle valuation.

Client
Cox Automotive
Services
UX/UI Design
App Development
3D Modeling
Big Data Analysis,
Labeling & Classification
ML Model Training
Tech Stack
Augmented Reality
Computer Vision
React Native
SQL
Python
Lidar
‍

Project overview

We partnered with Cox Automotive to tackle inconsistencies in vehicle inspections that occurred across various locations nationwide. Auto auctions rely on inspectors to evaluate vehicles based on factors like year, make, model, and condition. However, the absence of standardized training and the subjective nature of human inspections led to discrepancies in vehicle valuations. Cox Automotive needed a solution to ensure uniform assessments and consistent data.

To address this, we developed a robust mobile app using React Native, incorporating augmented reality (AR) and computer vision. The app allowed inspectors to analyze vehicles by overlaying real-time damage assessments on the car’s body panels via their device's camera. To power these capabilities, we developed three machine learning models, focusing on identifying the severity, location, and type of damage, enhancing the app’s precision. These models worked in tandem with computer vision algorithms to improve the accuracy of damage detection.

So, in a nutshell:
  • The client was experiencing subjective vehicle inspections, leading to inconsistent valuations across locations.
  • They lack standardized metrics for assessing the vehicle's damage severity and location.
  • Environmental factors were impacting the quality of inspections.
  • The limited ability to cross-reference vehicle condition against manufacturer models in real time.
    • Our solution:
  • The setup of a dedicated lookup database for updated costs in reported damages.
  • The development of a React Native, AR mobile app for real-time vehicle damage detection.
  • The training and implementation of 3 ML models to identify the type of damage, classify damage severity, and pinpoint location boosting the precision of the app’s computer vision capabilities.
  • Integrating AR overlays to help inspectors mark damage with high accuracy on vehicle body panels.
  • The design of a portable inspection tent prototype with lidar and 3D modeling to create a controlled, rapid inspection process for comparing vehicles against stock models for the next version of the tool.
    • Next Project