Next Best View Planning via Deep Reinforcement Leaning for 3D Reconstruction of Turbine Blades

The quality of 3D blade model reconstruction determines the accuracy of blade manufacturing. However, scanning equipment must gather data from multiple viewpoints to obtain a complete model due to the blade's complex shape and discontinuous surface. The choice of sensor views has a notable impact on the reconstruction task, commonly referred to as the nest-best-view (NBV) planning problem. The preferred approach generally involves either surface-based or volumetric methods. A deep reinforcement learning approach is proposed in this article to address the problem of determining the most suitable viewpoint. In addition, a virtual simulation model for 3D measuring system of turbine blades is built to acquire 3D measurement data and sensor pose. Based on this foundation, a deep reinforcement learning model is constructed and evaluated on two different shapes of blades datasets. The experiment's results showcase that the algorithm surpasses the baseline in the number of viewpoints.