Door Measurement with Computer Vision

Abstract

Abstract Artificial intelligence and computer vision have advanced considerably in recent years. In order to improve productivity and offer consumers ever more personalized products and services, many companies have developed applications for interacting, testing and visualizing their products. In this context, we were asked to explore various methods for measuring the dimensions of interior doors using computer vision. This functionality could be integrated into an application for previewing interior doors in a given setting. Previous work for this application consisted in removing the old door visible on an image in order to display a new door in 3D. The aim of this thesis is to examine different methods for measuring objects, in particular interior doors, using open source deep learning models and computer vision techniques. Thanks to these measurements, our application will be able to accurately position the new 3D door chosen by the user and estimate the dimensions required for a future order. At the end of this work, the method chosen was scale estimation from known objects. This method produced satisfactory results, with a relative error of less than 5 percents. It is also easy to use on a smartphone. However, it also has a drawback: the distance between the camera and the door when the image is taken. As the distance increases, the pixel quality of the photo decreases, which can lead to measurements with a relative error of over 5percents, making such measurements unacceptable.

Date of Award	26 Aug 2024
Original language	English
Awarding Institution	University of Namur
Supervisor	Patrick Heymans (Supervisor) & Ebrahim Khalil Abbasi (Co-Supervisor)