Computer vision for plastic bottle classification moving on conveyor

Abstract

Waste classification of used plastic bottles is one of the researches that have received much attention today. However, at present, this classification is still done manually by workers. In addition, there are also a number of studies that perform identification by traditional image processing methods. However, this does not bring about high identification efficiency because the shape of the plastic bottle is very complicated and it is also mixed in other types of waste. This paper will present a method to automatically recognize used water bottles when moving on a conveyor belt using machine learning techniques. This automatic identification method can assist workers in today's hard work of sorting waste. First, images of bottles are collected and preprocessed by conventional image processing methods, then these processed images are saved into a dataset for machine learning. The dataset is divided into 2 types of dirty bottles and clean bottles. The accuracy and processing time of recognition are verified experimentally. A 6-degree-of-freedom industrial manipulator with a camera fixed above the work area recognizes the coordinates of these plastic bottles and picks them up. Through experimental verification, accuracy of over 90% as well as identification time 200ms, the system proves to be applicable to the industrial environment of waste classification in the form of plastic bottles.