Publication
DNA-SLAM: Dense Noise Aware SLAM for ToF RGB-D Cameras
Oliver Wasenmüller; Mohammad Dawud Ansari; Didier Stricker
In: Asian Conference on Computer Vision Workshop. Asian Conference on Computer Vision Workshop (ACCV workshop-16), Taipeh, Taiwan, Province of China, Springer, 2016.
Abstract
SLAM with RGB-D cameras is a very active eld in Computer
Vision as well as Robotics. Dense methods using all depth and intensity
information showed best results in the past. However, usually they
were developed and evaluated with RGB-D cameras using Pattern Pro-
jection like the Kinect v1 or Xtion Pro. Recently, Time-of-Flight (ToF)
cameras like the Kinect v2 or Google Tango were released promising
higher quality. While the overall accuracy increases for these ToF cameras,
noisy pixels are introduced close to discontinuities, in the image
corners and on dark/glossy surfaces. These inaccuracies need to be specially
addressed for dense SLAM. Thus, we present a new Dense Noise
Aware SLAM (DNA-SLAM), which considers explicitly the noise characteristics
of ToF RGB-D cameras with a sophisticated weighting scheme.
In a rigorous evaluation on public benchmarks we show the superior
accuracy of our algorithm compared to the state-of-the-art.