Part-Based Room Categorization for Household Service Robots

A novel part-based image representation is proposed and an approach for room categorization using data obtained from a visual sensor is introduced. Images are represented with sets of unordered parts that are obtained by object-agnostic region proposals, and encoded using state-of-the-art image descriptor extractor - a convolutional neural network (CNN). An approach is proposed that learns category-specific discriminative parts for the part-based model. Outline of the room categorization method is depicted in Figure 1.

Figure 1: Outline of the part-based room categorization approach.

The proposed approach was compared to the state-of-the-art CNN trained specifically for place recognition. The baseline experiments demonstrate that both methods achieve comparable performance on original scene images, as shown in Table 1.

Table 1: Results of categorization using holistic CNN (top) and our part-based approach (bottom).

Further experiments revealed that our method outperforms the holistic CNN (Table 2) by being robust to image degradation, such as occlusions, modifications of image scaling, and aspect changes (Figure 1).

Figure 1: Examples of input image changes and deformations used in the robustness study.
Table 2: The overall categorization accuracy for experiments on original and distorted images using the holistic and part-based approach.

Publications for the topic of vision-based room categorization