The Power of GMMs: Unsupervised Dirt Spot Detection for Industrial Floor Cleaning Robots

Abstract

Small autonomous florr cleaning robots are the first robots to have entered our homes. These automatic vacuum cleaners have only used ver low-level dirt detection sensors and the vision systems have been constrained to plain-colored and simple-textured floors. However, for industrial applications, where efficiency and the quality of work are paramount, explicit high-level dirt detection is essential. To extend the usability of floor cleaning robots to theses real-world applications, we introduce a more general approach that detects dirt spots on single-colored as well as regularly-textured floors. Dirt detection is approached as a single-class classification problem, using unsupervised online learning of a Gaussian Mixture Model representing the floor pattern. An extensive evaluation shows that our method detects dirt spots on different floor types and that it outperforms state-of-the-art approaches especially for complex floor textures.

This work is supported by the European Commission through the Horizon 2020 Programme (H2020-ICT-2014-1, Grant agreement no: 645376).