Machine Vision for Solid Waste Detection

A huge variety of Municipal Solid Wastes (MSWs) caused by global urbanization and progress in development of packaging materials requires advanced sorting systems. Accurate and efficient recovery of recyclables from MSW, such as glass, paper, and plastics, is crucial to the development of a circular economy. Currently used optical solutions based near-infrared (NIR) and Visible Spectrum (VIS) enriched with machine learning methods for data analysis do not demonstrate adequate performance and efficiency in terms of solving detection and classification tasks. Still the sorting problem remains a complex set of tasks ranging from hardware and sensing to data analysis and inference. Moreover, the sorting process happens at the speed of around 6 m/s requiring perfect synchronization and high performance of individual units as well as the entire sorting system. In this chapter, we first start with the deep analysis of requirements and challenges in the scope of MSW automated sorting. Afterward we discuss hardware and sensing technologies serving as a data collection tool. We then cover the image processing topic stressing the following areas: image fusion and processing, computer vision methods, computer vision datasets, and data augmentation.