Visual and Location Information Fusion for Hierarchical Place Recognition

Abstract

Recognizing places of interest in an unfamiliar environment has been a common challenge faced by humans. This paper presents a novel hierarchical place recognition system capable of general outdoor place recognition including landmarks, commercial buildings, and business entities. We aim to achieve this by fusing visual and location information. Our hierarchical approach comprises place of interest detection, location-based filtering, image similarity score-based ranking and information retrieval components. The system leverages state of the art deep learning models for place detection and deep feature extraction. To evaluate our proposed system, we introduce a new dense dataset, referred to as Tokyo Outdoor Places, consisting of landmarks, commercial buildings, and business entities. Our proposed hierarchical system achieves 95.69% recall on our new dataset. We believe our system can contribute in achieving smart city goals by providing access to information, enabling locals and tourists to navigate with ease.