Search

The modern view of things in the science of robotics imposes that when working in a human environment, understanding of its equivalent semantics is required. In this paper, we present a graph-based unsupervised semantic clustering method and a novel cluster matching technique, with a view to create a multi-layer semantic memory map robust to illumination changes. Using indoor data collected by an unmanned aerial robot (UAR) and a publicly available dataset, we apply a community detection algorithm (CDA) to find efficiently coherent visual data throughout the trajectory creating a semantic base map. Then, we optimize the formed communities using metric information by implementing an hierarchical agglomerative clustering algorithm. The multilayer semantic map is created by constructing map instances for variant lighting conditions and matching the generated clusters to their base map correspondence. The proposed matching method relies on the graphs centrality indicators to identify central images of a region and utilize them to efficiently extract resemblances within the base map.

The paper at hand introduces a novel system for producing an enhanced semantic map that leverages a reconstruction approach of street-view scenes using computer vision and machine learning techniques. Focusing on the recognition and localization of objects/entities, the composed map combines semantic information from publicly available, yet of lower accuracy, satellite images, with more detailed data from ground-level camera measurements. This merging is achieved by utilizing odometry information from a street-moving vehicle and the 3D reconstruction of its recorded view. Then, the 3D semantic segmentation results are georeferenced and superimposed on the semantic map from the satellite images. In such a way, areas that require fine semantic accuracy can be improved, while the rest are left with the segmentation results of the satellite information. Every part of the proposed system is individually evaluated. We additionally test the overall approach on a case-study of georeferencing new labels of traffic signs, which are detected through a specifically designed classification network over a publicly available dataset collected around the city of Berlin.

Due to its vast applicability, the semantic interpretation of regions or entities increasingly attracts the attention of scholars within the robotics community. The paper at hand introduces a novel unsupervised technique to semantically identify the position of an autonomous agent in unknown environments. When the robot explores a certain path for the first time, community detection is achieved through graph-based segmentation. This allows the agent to semantically define its surroundings in future traverses even if the environment’s lighting conditions are changed. The proposed semantic clustering technique exploits the Louvain community detection algorithm, which constitutes a novel and efficient method for identifying groups of measurements with consistent similarity. The produced communities are combined with metric information, as provided by the robot’s odometry through a hierarchical agglomerative clustering method. The suggested algorithm is evaluated in indoors and outdoors datasets creating topological maps capable of assisting semantic localization. We demonstrate that the system categorizes the places correctly when the robot revisits an environment despite the possible lighting variation.

This work deals with the graph-based semantic segmentation of a robot’s traversed environment using the Louvain algorithm. In recent years, semantic segmentation has been the focus of several researchers’ interest and is applied to a variety of robotic applications. The Louvain method for community detection is a novel technique for extracting communities from large networks. The method is a greedy optimization one with a complexity of O(nlogn). We first assessed the Louvain algorithm with the COLD-Freiburg dataset to create the semantic map and compute the communities. We demonstrate that lighting conditions do not affect the system’s ability to categorize places. In particular, we train the system with COLD Freiburg seg1cloudy1 and we query images from seg1sunny1 and seg1night1. The results exhibit that the system is capable of categorizing the evaluated frames into the correct community.