The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLI-B5
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLI-B5, 857–863, 2016
https://doi.org/10.5194/isprs-archives-XLI-B5-857-2016
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLI-B5, 857–863, 2016
https://doi.org/10.5194/isprs-archives-XLI-B5-857-2016

  16 Jun 2016

16 Jun 2016

AUTOMATIC WATERLINE EXTRACTION FROM SMARTPHONE IMAGES

M. Kröhnert M. Kröhnert
  • Institute of Photogrammetry and Remote Sensing, Technische Universität Dresden, Germany

Keywords: Flood monitoring, Crowdsourcing, Smartphones, Image analysis, Texture measures

Abstract. Considering worldwide increasing and devastating flood events, the issue of flood defence and prediction becomes more and more important. Conventional methods for the observation of water levels, for instance gauging stations, provide reliable information. However, they are rather cost-expensive in purchase, installation and maintenance and hence mostly limited for monitoring large streams only. Thus, small rivers with noticeable increasing flood hazard risks are often neglected.

State-of-the-art smartphones with powerful camera systems may act as affordable, mobile measuring instruments. Reliable and effective image processing methods may allow the use of smartphone-taken images for mobile shoreline detection and thus for water level monitoring. The paper focuses on automatic methods for the determination of waterlines by spatio-temporal texture measures. Besides the considerable challenge of dealing with a wide range of smartphone cameras providing different hardware components, resolution, image quality and programming interfaces, there are several limits in mobile device processing power. For test purposes, an urban river in Dresden, Saxony was observed. The results show the potential of deriving the waterline with subpixel accuracy by a column-by-column four-parameter logistic regression and polynomial spline modelling. After a transformation into object space via suitable landmarks (which is not addressed in this paper), this corresponds to an accuracy in the order of a few centimetres when processing mobile device images taken from small rivers at typical distances.