The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLIII-B2-2021
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLIII-B2-2021, 433–440, 2021
https://doi.org/10.5194/isprs-archives-XLIII-B2-2021-433-2021
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLIII-B2-2021, 433–440, 2021
https://doi.org/10.5194/isprs-archives-XLIII-B2-2021-433-2021

  28 Jun 2021

28 Jun 2021

AUTOMATIC OBJECT SEGMENTATION TO SUPPORT CRISIS MANAGEMENT OF LARGE-SCALE EVENTS

S. M. Azimi1, R. Kiefl2, V. Gstaiger1, R. Bahmanyar1, N. Merkle1, C. Henry1, D. Rosenbaum1, and F. Kurz1 S. M. Azimi et al.
  • 1Remote Sensing Technology Institute, German Aerospace Center (DLR), Oberpfaffenhofen, Germany
  • 2German Remote Sensing Data Center, German Aerospace Center (DLR), Oberpfaffenhofen, Germany

Keywords: Crisis Management, Segmentation, Aerial Imagery, Large-scale Events, Machine Learning

Abstract. The management of large-scale events with a widely distributed camping area is a special challenge for organisers and security forces and requires both comprehensive preparation and attentive monitoring to ensure the safety of the participants. Crucial to this is the availability of up-to-date situational information, e.g. from remote sensing data. In particular, information on the number and distribution of people is important in the event of a crisis in order to be able to react quickly and effectively manage the corresponding rescue and supply logistics. One way to estimate the number of persons especially at night is to classify the type and size of objects such as tents and vehicles on site and to distinguish between objects with and without a sleeping function. In order to make this information available in a timely manner, an automated situation assessment is required. In this work, we have prepared the first high-quality dataset in order to address the aforementioned challenge which contains aerial images over a large-scale festival of different dates. We investigate the feasibility of this task using Convolutional Neural Networks for instance-wise semantic segmentation and carry out several experiments using the Mask-RCNN algorithm and evaluate the results. Results are promising and indicate the possibility of function-based tent classification as a proof-of-concept. The results and thereof discussions can pave the way for future developments and investigations.