Image-based house furniture price estimation

Published by Samy Doloris on

Artificial Intelligence has been used in insurance for almost as early as the field existed. From fraud detection to customer behavior modeling (profiling), or more conventional applications such as forecasting, AI always found a way to make its place in this industry.

A business value that AI can bring and which is often misregarded in that sector is user experience. Indeed, AI can be used to automate recurrent tasks and processes.

When we got to work on Flora insurance, the goal was pretty clear: design a new insurance product, fully digitized and with a real user experience.

In that sense, we thought about using AI for image-based price estimation, and I will explain how!

YOLO (You Only Look Once)

YOLO is a Deep Learning model that allows extremely fast and accurate object detection.

As you can see on the video below (featuring our data science team 😉 ), it works pretty well on videos too!

This video was taken with my laptop (CPU: Intel i5-8265U and no GPU) and processed afterwards. The video was processed at around 1 or 2 frames per second, which is a relatively impressive performance of the model on such a laptop without a GPU (a good GPU allows real-time processing).

Goal

The goal of using such a model is to detect house furniture on picture or videos and, based on that information, estimate the price of the content that can be seen.
The approach is to maintain a price table for each detectable object, as it can be seen on the figure below:

An image is passed to YOLO. The model outputs a list of detected objects as well as their positions (bounding boxes), and confidence levels (probability) for each detected object.

The output is compared to the price table and a summary report is generated.

Business value

As mentioned in the introduction, the main business value of our solution lies in enhancing the user experience:

  • The customer has at his/her disposal a new and easier way to encode the content he/she wants to insure.
  • The possibility to edit the detected content is given. It allows to correct and enhance the model itself.
  • Personalized offers and summary reports (business intelligence) can be implemented easily.
  • Enhancing the user experience allows to reach a new target market by building a fully digitized insurance product.

Stay tuned for more information about the solution and the product it belongs to!