Environment Sensing and Landmarks Detection for Accurate Self-Localisation

Abstract

In the future system, self-localization emerges as a critical approach to address various challenges, enhance accuracy, and alleviate network-side burdens by shifting computational load to terminal devices. Leveraging diverse sensing and machine learning capabilities becomes instrumental in facilitating self-localization, especially when terminals are equipped with sensors like Light Detection and Ranging (LiDAR) alongside limited computational resources. This paper introduces a self-localization method wherein terminals detect landmarks with known locations in the environment to estimate their own positions. Using LiDAR, terminals scan the surroundings and employ a pretrained machine learning model to detect landmarks. Subsequently, they estimate their positions by comparing the perceived locations of landmarks (derived from LiDAR measurements) with their actual positions. An experiment conducted in our laboratory utilized regular shapes to represent landmarks in any environment, such as doors, windows, and furniture. The results demonstrate accurate location estimation, with errors less than 10 cm when a single landmark is detected. The single-landmark localisation is particularly beneficial in dense environments (Line-of-Sight availability to landmarks is limited) or those with limited features. Additionally, performance significantly improves when multiple landmarks are detected, showcasing the robustness and effectiveness of the proposed approach. The results also show that performance is improved significantly and the error is less than 3 cm when multiple landmarks are detected.