14 October 2024 in IEEE Transactions on Instrumentation and Measurement
Context-adaptive navigation (CAN) systems enable seamless navigation in complex environments by identifying the surrounding contexts of ground vehicles. Developing a fast, reliable, and robust navigational context recognition (NCR) method is crucial for the effective operation of CAN systems. However, most existing context recognition approaches focus on coarse binary or quaternary indoor/outdoor detection, which is not tailored to the specific needs of CAN applications. In this paper, we propose a fine-grained context categorization framework consisting of seven environment categories for vehicle navigation, representing the most elaborate context categorization framework in this research domain. To address the inter-class confusion resulting from the increased number of categories, a novel feature is devised by utilizing the carrier-power to noise-density ratio ( C / N 0) and azimuth derived from Global Navigation Satellite System (GNSS) measurements. Additionally, a low-complexity algorithm is developed to extract this feature efficiently. Furthermore, to ensure real-time performance, we employ a lightweight gated recurrent unit (GRU) network for its superior sequence data processing capabilities. A dataset containing 120,436 samples is created and made publicly available. Extensive experiments conducted on this dataset demonstrate that our proposed method achieves an overall recognition accuracy of 99.43% for isolated scenarios and 95.24% for transition scenarios, with an average transition delay of 1.37 seconds.