Accelerated Reconstruction of Scenes Using CUDA-Based Parallel Computing

In recent years, perceiving the environment has been the most important part of unmanned aerial vehicles (UAVs) inspection tasks on the battlefield scene. This article aims to study how to quickly reconstruct unknown battlefield scenes, such as buildings, streets, etc. using UAVs equipped with binocular cameras and stereo-matching technology. We use a binocular module to reduce the cost of drone-borne equipment and lighten the load of the UAVs. Since the drone needs to use an embedded onboard computer and stereo matching algorithms require massive computing resources, this paper studies the optimization of CUDA-based stereo matching algorithms on embedded devices to improve the reconstruction effect of scenes and achieve rapid reconstruction, thereby reducing time costs, and enhancing drone efficiency.