Indoor vs. Outdoor Scene Classification for Mobile Robots.

This paper deals with the task of automatic indoor vs. outdoor classification from image data with respect to future usage in mobile robotics. For the requirements of this research, we utilize the Mini-places dataset. We compare a large number of classic machine learning approaches such as Support Vector Machine, k-Nearest Neighbor, Decision Tree, or Naive Bayes using various color and texture description methods on a single dataset. Moreover, we employ some of the most important neural network-based approaches from the last four years. The best tested approach reaches 96.17% classification accuracy. To our best knowledge, this paper presents the most extensive comparison of classification approaches in the task of indoor vs. outdoor classification ever done on a single dataset. We also address the processing time problem, and we discuss using the applied methods in real-time robotic tasks.