Military simulation applications put strong requirements on terrain modelling. Large mission and training areas need to be represented in detail, while users expect these models to be available at ever shorter lead times. Capabilities are needed that rapidly transform sensor data into models that fully represent the complexity of the mission environment. Industry seems to have solved part of the problem with mature photogrammetric techniques and LiDAR data acquisition. However, the data delivered by these techniques is often limited to a geometric and only visual model that has little semantics and as such is not ready for simulation.
Current sensor data analysis techniques result in labeled imagery and point clouds, assigning semantics to pixels and points. At best, the points are then converted into semantic linears or areals. The challenge is to find complete models that match with the geometry and semantics of the points. The research presented in this paper addresses this challenge. We seek techniques that directly extract semantically rich and simulation-ready models from sensor data. Our hypothesis is that procedural modelling techniques are key to the solution and that innovative application of modern data analysis techniques is required to delve instances of these procedural models from the sensor data. We introduce the concept of model mining to refer to the process that finds these models by fitting optimized models to sensor data.
In our paper we report on results we have achieved with model mining applied to a drone based point cloud dataset. We use particle swarming optimization techniques to find procedural models within the data. Model mining is a complex problem that needs extensive research to mature. We hope this paper will trigger others to take the challenge of model mining and bring rapid terrain analysis to the military simulation community.
Keywords
AI, GEOSPATIAL DATA , MODELING, RAPID MODELING, TERRAIN
Additional Keywords