Real Time 3D Segmentation System Based on YOLOv8 and SAM for Sorting Recyclable Plastic Bottle Waste

The abundance of plastic products around the world is causing social, environmental, and economic problems. Many countries are trying to recycle plastic waste, but the recycling rate is very low. In the case of plastic PET products, it is valuable to recycle economically and environmentally because the recycling rate and quality are higher than other plastic materials. This paper proposes a real-time 3D segmentation system based on the YOLOv8 and segment anything model (SAM) to classify plastic bottles being discarded along with other types of waste. The proposed system basically uses a stereo camera to obtain waste images and depth information, and goes through a preprocessing procedure. The preprocessed data is input into YOLOv8 to detect plastic bottles. Then, the detected bottle image and bounding box information are input to SAM to achieve meaningful segmentation, and then a 3D segment is obtained along with depth information. Finally, the system prioritizes the detected bottles with weights, which are utilized to pick up in the segregation robot. This paper also describes the experimental results and performance evaluation to verify the effectiveness of the proposed method.