Locating and identifying furniture elements using computer vision
A new furniture system designed for the TU Delft library, featuring a range of unique and flexible work and relaxation spaces inside three-dimensional voronoi cells. The system includes removable cells that can be placed throughout the library space. To place the abstract shapes back in the structure, they need to be identified. We've developed an object detection model to facilitate this process, which we're showcasing in this submission. We trained our computer vision model using rendered images. The result is an efficient and accurate model that can identify and locate the removable cells in real-time. If you're interested in seeing our object detection model in action, we invite you to watch our accompanying video.
This project presents a new furniture system designed specifically for the TU Delft library, incorporating a range of unique and flexible work and relaxation spaces inside three-dimensional voronoi cells. The orange colored cells are removable and can be placed throughout the library space, offering users a variety of different settings to work and relax in. To reattach the cells to the main structure, their shapes need to be identified. To facilitate this process, we developed an object detection model that uses computer vision to identify and locate the removable cells in real-time. The model was trained using rendered images, where both the camera pos- ition and the position of all cells was randomized for every training image. We based our model on the YOLOv5 open source model. The YOLOv5 model is a state-of-the-art computer vision system that provides real-time object detection and tracking capabilities. The resulting model is intended to be used by users on their smartphones, enabling them to walk through the library space and use the camera of their phone to identify the removable cells and show the user where they should be placed back into the structure. We created an accompanying video that shows a realtime preview of the model.
Team members : Oliver J. Post, Jiacheng Xu, Jiahui Shi
Supervisor : H. Bier, A. Hidding, S. Khademi, C. van Engelenburg, A. Luna Navaro, S. Brancart, V. Laszlo
Institution : TU Delft
Oliver J. Post