2018/06/27 by Benjamin Hepp, Hepp, Benjamin, Debadeepta Dey +9 · 1 citation
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.1806.10354
openalex publication_date 2018/06/27 · openalex created_date 2022/10/02 · openalex updated_date 2026/07/28
Camera equipped drones are nowadays being used to explore large scenes and\nreconstruct detailed 3D maps. When free space in the scene is approximately\nknown, an offline planner can generate optimal plans to efficiently explore the\nscene. However, for exploring unknown scenes, the planner must predict and\nmaximize usefulness of where to go on the fly. Traditionally, this has been\nachieved using handcrafted utility functions. We propose to learn a better\nutility function that predicts the usefulness of future viewpoints. Our learned\nutility function is based on a 3D convolutional neural network. This network\ntakes as input a novel volumetric scene representation that implicitly captures\npreviously visited viewpoints and generalizes to new scenes. We evaluate our\nmethod on several large 3D models of urban scenes using simulated depth\ncameras. We show that our method outperforms existing utility measures in terms\nof reconstruction performance and is robust to sensor noise.\n