Categorization of indoor places by combining local binary pattern histograms of range and reflectance data from laser range finders

Abstract

This paper presents an approach to categorize typical places in indoor environments using 3D scans provided by a laser range finder. Examples of such places are offices, laboratories, or kitchens.In our method, we combine the range and reflectance data from the laser scan for the final categorization of places. Range and reflectance images are transformed into histograms of local binary patterns and combined into a single feature vector. This vector is later classified using support vector machines. The results of the presented experiments demonstrate the capability of our technique to categorize indoor places with high accuracy. We also show that the combination of range and reflectance information improves the final categorization results in comparison with a single modality.

Semi-supervised 3D place categorisation by descriptor clustering

“…KyushuIndoor is the only relevant data set for 3D place categorisation from the literature that we are aware of. By comparing our results from clustering of NDT histogram descriptors to two baselines -the state-of-the-art 3D SVM classifier of Mozos et al [10] and clustering with a 2D descriptor used in previous works [12] -we show that k-means clustering of the NDT appearance descriptor attains high accuracy without training. By employing hierarchical k-means++, we have the added benefit of a semantically meaningful sub-categorisation of places.…”

“…III and IV to three data sets: the office-like benchmark set used in Mozos et al [10] (KyushuIndoor), one from a warehouse (ArlaWarehouse), and one from an outdoor field robot (EskilstunaField).…”

“…Pronobis et al [13] extend this work by combining laser and camera, and training a support vector machine (SVM) to combine image and range features. In later work, Mozos et al [10,11] use 3D scans. This work is particularly relevant for the present paper, and we use the same 3D data set [10] in our main quantitative results (Sec.…”

“…In later work, Mozos et al [10,11] use 3D scans. This work is particularly relevant for the present paper, and we use the same 3D data set [10] in our main quantitative results (Sec. V-B, Tab.…”

“…Place categorisation may improve service robots' communication capabilities with humans [5,10], and holds a central place in semantic mapping. This ability may be particularly useful for teleoperation, so that the robot can tell operators what environment it is in, when that is hard to see from a video stream alone.…”