2020/11/03 by René Schuster, Schuster, René, Christian Unger +3
Computer Science · #Advanced Image Processing Techniques #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Enhancement Techniques
paper · pdf · doi:10.48550/arxiv.2011.01603
openalex publication_date 2020/11/03 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Motion estimation is one of the core challenges in computer vision. With\ntraditional dual-frame approaches, occlusions and out-of-view motions are a\nlimiting factor, especially in the context of environmental perception for\nvehicles due to the large (ego-) motion of objects. Our work proposes a novel\ndata-driven approach for temporal fusion of scene flow estimates in a\nmulti-frame setup to overcome the issue of occlusion. Contrary to most previous\nmethods, we do not rely on a constant motion model, but instead learn a generic\ntemporal relation of motion from data. In a second step, a neural network\ncombines bi-directional scene flow estimates from a common reference frame,\nyielding a refined estimate and a natural byproduct of occlusion masks. This\nway, our approach provides a fast multi-frame extension for a variety of scene\nflow estimators, which outperforms the underlying dual-frame approaches.\n