The article discusses the use of a dynamically weighted factor graph model to improve feature-based geo-localization, which involves associating features from aerial imagery with those detected by a vehicle’s sensors. The model adjusts the weight based on information from the LiDAR sensor detections and includes a prior error estimation. The model dynamically adjusts weights when the representation becomes ambiguous or sparse, reducing outliers and deviations. The authors compare their method with other geo-localization methods in a challenging ambiguous environment, demonstrating its effectiveness in mitigating the mentioned drawbacks.
Publication date: 14 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.07301
