VibeBin

Abstract

This paper presents the design and implementation of VibeBin, a low-cost, non-intrusive and easy-to-install waste bin level detection system. Recent popularity of Internet-of-Things (IoT) sensors has brought us unprecedented opportunities to enable a variety of new services for monitoring and controlling smart buildings. Indoor waste management is crucial to a healthy environment in smart buildings. Measuring the waste bin fill-level helps building operators schedule garbage collection more responsively and optimize the quantity and location of waste bins. Existing systems focus on directly and intrusively measuring the physical quantities of the garbage (weight, height, volume, etc.) or its appearance (image), and therefore require careful installation, laborious calibration or labeling, and can be costly. Our system indirectly measures fill-level by sensing the changes in motor-induced vibration characteristics on the outside surface of waste bins. VibeBin exploits the physical nature of vibration resonance of the waste bin and the garbage within, and learns the vibration features of different fill-levels through a few garbage collection (emptying) cycles in a completely unsupervised manner. VibeBin identifies vibration features of different fill-levels by clustering historical vibration samples based on a custom distance metric which measures the dissimilarity between two samples. We deploy our system on eight waste bins of different types and sizes, and show that under normal usage and real waste, it can deliver accurate level measurements after just 3 garbage collection cycles. The average F-score (harmonic mean of precision and recall) of measuring empty, half, and full levels achieves 0.912. A two-week deployment also shows that the false positive and false negative events are satisfactorily rare.