RGB-D SLAM with Deep Depth Completion

SERHATOĞLU A. O., Guclu O., CAN A. B.

21st International Conference on Artificial Intelligence and Soft Computing, ICAISC 2022, Zakopane, Polonya, 19 - 23 Haziran 2022, cilt.13589 LNAI, ss.59-67 identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 13589 LNAI
  • Doi Numarası: 10.1007/978-3-031-23480-4_5
  • Basıldığı Şehir: Zakopane
  • Basıldığı Ülke: Polonya
  • Sayfa Sayıları: ss.59-67
  • Anahtar Kelimeler: Depth completion, Depth estimation, SLAM
  • Hacettepe Üniversitesi Adresli: Evet

Özet

RGB-D indoor mapping has been an active research topic in the last decade with the release of various depth sensors. Researchers proposed impressive SLAM systems such as ORB-SLAM2. However, the depth sensors are sensitive to illumination conditions and have limited range, which lead to missing or invalid depth data. This situation negatively affects the performance of RGB-D SLAM systems. Moreover, deep learning based approaches for estimating depth data from color frames have been proposed recently. Therefore, in this study, we aim to analyze deep depth estimation performance on SLAM. We propose a depth completion approach which merges sensor depth data and the estimated depth. To do this, we integrate a deep depth estimation method into a state-of-the-art indoor RGB-D SLAM system. The experimental results show that the proposed depth completion approach improves mapping performance.