Object detection with YOLO for the low-cost educational robot E-puck 2
Supervisors: Julien Nembrini
Student: Vincent Carrel
Project status: Finished
Year: 2022
At the University of Fribourg, the students following the robotics course are tasked with doing their own personal challenge, which were usually limited by the difficulty of working with the camera. In this project, an object detection module is added to the UNIFR API EPUCK, a Python API developed by David Frischer in the context of another Bachelor Thesis, used to control e-puck2, which are small educational robots. The new methods will give informations (position, class, confidence) about different object in the field of view of the robot. The field of objects detection in general is discussed. Then the YOLO algorithm (You Only Look Once) is explained in detail. A training data set is created, annotated and used to train the model with real situations. The model is implemented as an add-on, combined with the current API, with easy to use methods for future students. The quality of the model is tested with a benchmark, comparing the results of multiple robots in different situations and environments, including a new set-up which was not present in the training data set.
Document: report.pdf