Dynamic Wind Turbine Blade 3D Model Reconstruction with Event Camera

In this paper, we proposed E2DSNeRF, a new image-based method that is highly suitable for 3D reconstruction of wind turbines. Specifically, considering the smooth and textureless characteristics of wind turbine blades that can lead to blurring and ghosting, we use event cameras with high dynamic range and low latency instead of regular RGB-D cameras to capture images of wind turbines. Then, compared to the dozens of images required by the vanilla neural radiation field method NeRF, we used very sparse input views, which greatly reduced the burden of image acquisition and preprocessing. In addition, the use of deep supervision not only improves the rendering quality of sparse input views, but also eliminates blurring and ghosting caused by low textures on wind turbine blades. At the same time, in order to obtain a more accurate camera pose, we added necessary feature points to the background when collecting the wind turbine dataset to ensure that the camera pose obtained by COLMAP is sufficiently accurate. Due to sparse input views, our method has significantly improved training speed compared to NeRF. The experimental results demonstrate the feasibility and effectiveness of this method, which has lower cost and clearer reconstruction results compared to traditional methods.