An experimental evaluation of Siamese Neural Networks for robot localization using omnidirectional imaging in indoor environments

TL;DR

This work investigates robot localization from omnidirectional panoramic images using Siamese CNNs to learn robust global appearance descriptors. It tackles two tasks—room discrimination and global localization—by formulating them as image-pair similarity problems and image retrieval within a map, respectively, and evaluates multiple CNN backbones and descriptor sizes with a contrastive loss. Data augmentation to simulate illumination changes and rotations improves robustness, yielding competitive localization accuracy especially under cloudy and night conditions on the COLD-Freiburg dataset. The findings demonstrate that careful choices of feature extractors and descriptor dimensions enable effective place recognition and pose estimation in indoor, panoramic imaging scenarios, with clear guidance for practical deployments and future extensions to outdoors and multi-sensor fusion.

Abstract

The objective of this paper is to address the localization problem using omnidirectional images captured by a catadioptric vision system mounted on the robot. For this purpose, we explore the potential of Siamese Neural Networks for modeling indoor environments using panoramic images as the unique source of information. Siamese Neural Networks are characterized by their ability to generate a similarity function between two input data, in this case, between two panoramic images. In this study, Siamese Neural Networks composed of two Convolutional Neural Networks (CNNs) are used. The output of each CNN is a descriptor which is used to characterize each image. The dissimilarity of the images is computed by measuring the distance between these descriptors. This fact makes Siamese Neural Networks particularly suitable to perform image retrieval tasks. First, we evaluate an initial task strongly related to localization that consists in detecting whether two images have been captured in the same or in different rooms. Next, we assess Siamese Neural Networks in the context of a global localization problem. The results outperform previous techniques for solving the localization task using the COLD-Freiburg dataset, in a variety of lighting conditions, specially when using images captured in cloudy and night conditions.

Figures (3)

• Figure 1: Representation of the architecture of a general Siamese Neural Network.
• Figure 2: Detailed representation of a Siamese Neural Network with AlexNet in the feature learning phase and the flattening phase composed of three fully connected layers.
• Figure 5: Example of different trajectories of the robot.