基於視覺與無線網路融合的室內環境穩健SLAM 系統

Abstract

視覺 SLAM（同時定位與地圖構建）已被廣泛應用於室內機器人或是自駕車領域中。然而，在現實世界的室內環境中，視覺 SLAM 可能因為存在動態物體、照明變化或缺乏特徵等原因而無法正常工作，進而導致機器人位置估計不準確或跟踪失敗。此外，不同位置的特徵類似可能會導致錯誤的回環檢測以及回環校正。 由於大多數現代室內環境已經具備 WiFi 基礎設施，WiFi 定位另一種常用於室內定位的技術。然而，大多數基於接收信號強度指示（RSSI）的 WiFi 定位方法需要離線數據庫構建，這在不需要先驗知識的 SLAM 應用情景下並不適用。 為了解決這些問題，本論文提出了一種能夠實時運行並交互更新 WiFi 和視覺訊息的方法，提供 WiFi 和視覺 SLAM 相結合的解決方案，通過在視覺 SLAM 加入 WiFi 訊息，該系統在環境中更具魯棒性，同時提供了一種具有成本效益的解決方案。

Visual SLAM (Simultaneous Localization and Mapping) has been widely utilized for indoor service robots. However, in real­world indoor environments, Visual SLAM can encounter issues that hinder its proper functioning. These issues include the presence of dynamic objects, changes in illumination, or a lack of discernible features, leading to inaccurate estimation of the robot’s position or tracking failure. Additionally, perceptual aliasing can arise when different locations exhibit similar characteristics, resulting in false loop closure detection. WiFi localization is another popular technology employed for indoor positioning, as most modern indoor environments are already equipped with WiFi infrastructure. However, many WiFi localization methods based on Received Signal Strength Indication (RSSI) necessitate offline database construction, which is unsuitable for SLAM use cases that operate without prior knowledge. To tackle these challenges, this thesis proposes a method capable of running in realtime, actively updating both WiFi and vision information. This method integrates WiFi and visual SLAM to enhance the visual SLAM system using WiFi signals. By leveraging WiFi information, the system becomes more robust to the environment and offers a costeffective solution.