A Survey on Visual Place Recognition for Mobile Robots Localization

Abstract

Visual place recognition is an active research field in the robotic navigation and localization, which means the ability to recognize a known place in the environment using vision as the main sensor modality. Despite significant progress in computer vision and machine learning techniques, challenges remain especially in dynamic environments such as illumination change, viewpoint change and so on. In this paper, a survey and comparative study on existing approaches of visual place recognition is presented, including place feature extraction methods, image similarity metrics and searching algorithms, as well as some benchmark datasets and evaluation metrics. Experimental results show that the methods combining feature extraction using convolutional neural networks and sequential image searching achieve higher precision in large scale dynamic environment.