UACE2017 Proceedings: Physically based sonar simulation and image generation for side-scan sonar

• Session:
  Signal and image processing
• Paper:
  Physically based sonar simulation and image generation for side-scan sonar
• Author(s):
  Philipp Woock, Jürgen Beyerer
• Abstract:
  The paper describes a method to generate side-scan sonar images from any geometry and vehicle path. These generated images can be used for evaluation of reconstruction algorithms or for training purposes (be it human or for deep neural networks).\nTo achieve this, we use the open source modeling tool Blender in conjunction with the Mitsuba renderer and a custom sonar signal processing stage.\nThe basic workflow is to model any geometry and vehicle path in Blender. The geometry information is then exported to a scene description format which the Mitsuba renderer is able to read. From there, we let Mitsuba perform raytracing on the scene and obtain various scene maps: per-pixel distances to the surface elements in the scene and per-pixel information about the normal vector of the surface element. \nGiven that information, a physically based sonar simulation is done by modeling the travel of individual sound packets to the surface elements. This simulation can be done in a multitude of ways: For chirped sonars or continous-frequency sonars and it can be done coherently or incoherently. An available sonar beam pattern is also taken into account in the simulation to correctly model the narrow beam of a side-scan sonar. Due to the generic modelling, the simulation is not restricted to a single device and realistic measurements can be simulated as the generated images exhibit the typical speckle noise characteristic which is a problem in many applications.
•   Download the full paper

Contact details

• Contact person:
  Dr Philipp Woock
• e-mail:
  
• Affiliation:
  Fraunhofer IOSB
• Country:
  Germany