Multi-scale Vehicle Localization in Underground Parking Lots by Integration of Dead Reckoning, Wi-Fi and Vision

摘要

In this paper, we present an improved vehicle localization scheme for underground parking lots. We are integrating Dead Reckoning techniques, Wi-Fi signals, and vision-based localization methods. the proposed method includes multiple offline and online localization stages. In the offline stage, the Wi-Fi data and pavement images are collected, whereas, in the online stage, multi-scale localization estimation is performed with three stages. First, we derive a coarse localization result by integrated the Dead Reckoning (DR) technique, and Wi-Fi fingerprint localization is used to detect direction change in the underground parking lot. Second Image level localization, the involves the application of pavement image matching by Oriented FAST and Rotated Binary Robust Independent Elementary Features BRIEF methods (ORB). Finally, the Random sample consensus algorithm (RANSAC) is used for removing false points of coincidence and improved efficiency. The proposed solution achieves a 96% success rate with localization errors averaging under one meter.