Optimizing Food Sample Handling and Placement Pattern Recognition with YOLO: Advanced Techniques in Robotic Object Detection

Abstract

Japan, grappling with a significant labor deficit driven by a declining birthrate and an aging population, is increasingly turning towards industrial robotics, particularly in the food manufacturing sector. The diverse nature of food products presents a notable challenge: the need for customized programming for each product variant, which has been a major impediment to the widespread adoption of robotics in this domain. This paper presents an innovative solution utilizing the "You Only Look Once" (YOLO) object detection algorithm, aimed at simplifying the programming process for industrial robots. A key enhancement in our approach is the integration of Non-Maximum Suppression (NMS) for effective discrimination of overlapping objects, leveraging critical data such as the center of gravity, depth, and bounding box measurements. The incorporation of RGB-D sensors enables the acquisition of precise spatial height data, essential for assessing the stacking and arrangement of items. The practicality and effectiveness of this methodology are corroborated through empirical trials involving the robotic placement of real-world objects. This research not only demonstrates the feasibility of our approach but also underscores its potential to significantly optimize robotic operations in food manufacturing.