Place categorization through object classification

This paper proposes a novel methodology for place categorization in mobile robots based on the presence of objects. In order to achieve such categorization, the robot is equipped with an RGB-D sensor. For a given time interval the sensor's measurements are combined with robot's localization data and reconstruct the 3D scene from the respective pointclouds. Afterwards, the method searches for dominant planes which are the most probable locations for finding objects. Given those planes, this work seeks and discriminates objects. The recognized objects, form a distribution which is given as input to a Naive Bayesian classifier in order to categorize the place.