Real-Time Pieces Detection and Recognition by the NAO Robot Using Tiny-YOLOv4/3L Algorithm

This research paper presents a system for object detection using the Tiny-YOLOv4/3L model, with a focus on a humanoid robot NAO. The system is designed to detect specific objects under various conditions. The efficacy of the model was evaluated based on its performance under standard conditions and under common real-world scenarios such as occlusion and low light. The results demonstrate that the model can provide reliable detection with commendable accuracy under standard conditions and shows a degree of resilience against minor occlusion and low light. However, significant occlusion and dim light conditions pose considerable challenges to the performance of the model.