OfficeVibe Dataset: Office Activity Classification through Human-Induced Structural Vibrations

We present OfficeVibe, a structural vibration dataset collected from 10 people performing office activities at 2 different locations. Recognizing human activities in an office has great potential for productivity tracking and health monitoring applications. OfficeVibe is an ambient structural vibration dataset for human activity recognition in office environments. The vibration-based approach preserves privacy because it does not preserve speech or images or understand what is typed or written. It may be possible to identify people based on the characteristics of the vibration caused by their activities, but this would require pre-training on the specific people we want to identify, which is easy to avoid. The intuition of the vibration-based office activity recognition method is that human motions (macro and micro motions) induce the ambient structures to vibrate. The vibration wave travels through the structures (e.g. tables, floor) and does not require line-of-sight to propagate. OfficeVibe has labels on office activities (QUIET, WRITE, TALK, and TYPE), and (SIT, STAND, WALK). This allows us to train classifiers that recognize specific activities that may be rare without overfitting the available training data. This dataset aims to explore and characterize vibration-based activity recognition at locations where IoT sensors are likely to be placed (e.g. tables, floors, shelves).