Image recognition reduced compliance errors by 30%

Context

A leading French risk management consulting firm annually certifies the electrical compliance of buildings.

The Challenge

The firm's technicians rely on paper reports from the previous year to verify compliance. This process is time-consuming and prone to errors.

Our mission was to develop a decision support tool that automatically identifies differences between two photos of the same electrical panel taken one year apart.

The Solution

The solution consists of two steps. First, we realigned the photos using the SIFT algorithm. Then, we compared the photos using the algorithms of the OpenCV framework.

We focused on reducing false positives, even if it meant ignoring some real cases. This approach often promotes user adoption of the solution.

The Results

We built the application in 3 months and complete the originally set objectives.