A Uniform Distribution of Landmarks for Efficient Map Compression

In this paper, we address the challenge of visual-based localization in dynamic outdoor environments characterized by continuous appearance changes. These changes greatly affect the visual information of the scene, resulting in significant performance degradation in visual localization. The issue arises from the difficulty of mapping data between the current image and the landmarks on the map due to environmental variations. One approach to tackle this problem is continuously adding new landmarks to the map to accommodate diverse environmental conditions. However, this leads to map growth, which in turn incurs high costs and resource demands for localization. To address this, we propose a map management approach based on an extension of the state-of-the-art technique called Summary Maps. Our approach employs a scoring policy that assigns scores to landmarks based on their appearance in multiple localization sessions. Consequently, landmarks observed in multiple sessions are assigned higher scores. We demonstrate the necessity of maintaining landmark diversity throughout map compression to ensure reliable long-term localization. To evaluate our approach, we conducted experiments on a dataset comprising over 100 sequences encompassing various environmental conditions. The obtained results were compared with those of the state-of-the-art approach, showcasing the effectiveness and superiority of our proposed method.