Automatic context learning based on 360 imageries triangulation and 3D LiDAR validation

Geographic data is very valuable for decision making. There are many hand-adapted datasets of roads or buildings available. However, datasets of other objects are not available, and it is very difficult to generate them manually. Remote sensing can help us to generate datasets of specific objects. This work introduces the main components for an automatic dataset generation process using any kind of sensors. To validate this process, an implementation using an open-source dataset is developed, geolocating traffic barriers using 360-degrees images captured from a car. Its results are validated with the positions extracted from a 3D LiDAR, solving the same problem at a much lower cost, providing an acceptable error for some use cases.