Abstract
In this paper, an efficient normal estimation and filtering method for depth images acquired by Time-of-Flight (ToF) cameras is proposed. The method is based on a common feature pyramid networks (FPN) architecture. The normal estimation method is called ToFNest, and the filtering method ToFClean. Both of these low-level 3D point cloud processing methods start from the 2D depth images, projecting the measured data into the 3D space and computing a task-specific loss function. Despite the simplicity, the methods prove to be efficient in terms of robustness and runtime. In order to validate the methods, extensive evaluations on public and custom datasets were performed. Compared with the state-of-the-art methods, the ToFNest and ToFClean algorithms are faster by an order of magnitude without losing precision on public datasets.
Highlights
With the evolution of 3D sensors in the robotics domain, the focus in the perception shifted towards the 3D point cloud-based sensing
The normal estimation of the point clouds has a long history in the research community, with publications on this topic dating back a long time [21], and it is still a hot topic with a considerable amount of publications in the last year focusing on learning-based methods [13,16,18,22]
The method itself is generic to accept additional information besides the depth image: the input consists of three levels, for which one is used for point cloud normal estimation, while the two additional layers enable encoding camera-specific information such as IR
Summary
With the evolution of 3D sensors in the robotics domain, the focus in the perception shifted towards the 3D point cloud-based sensing. The level of noise that corrupts the 3D data is one of the primary problems of normal estimation [7,8] This can be reduced with advanced filtering methods, which are relevant for other components of a point cloud processing pipeline [9]. The idea of adaptive scale is present for the Feature Pyramid Networks, which we adopted in our approach with the insight that they are able to mimic a multi-scale behaviour These methods showed that a considerable enhancement can be achieved by using recent deep learning-based techniques for point clouds [12,13,14,15,16,17,18,19].
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
