Studying behaviour with video analysis tools

Designing infrastructure, we have in mind some ideas of how it is going to be used and what behaviour and interactions between road users it will promote. However, only through observing the actual behaviour we can prove whether our ideas were correct or not. Ideally, we would like to have a set of objective and measurable indicators that can be collected (observed) in traffic without interrupting its normal functioning, and be able to relate them to qualities like safety and efficiency of the traffic system. Having this in hands, we can quantify how well the infrastructure design succeeds in promoting the desired traffic qualities and whether changes in the design are necessary. One of the technologies that have potential to greatly aid the collection of the behavioural indicators in traffic is video analysis. Depending on the complexity of the algorithms used, it is possible to detect simple events such as the presence of a road user in a certain area, simultaneous arrivals, extract trajectories and speed profiles for individual road users and calculate various indicators that describe the process of their interaction. The other side of the coin is, the more advanced the algorithm is, the more efforts it requires for calibrating the intrinsic parameters, the more sensitive it is to changes in the scene, viewing angle, lighting conditions, and the more computer power it requires. A number of video analysis tools for traffic applications exists, however, except for those performing the most basic tasks like counts or presence detection, none of them is ready to be handed over to practitioners for regular use. The main challenge is the complexity of the problem to solve when road users of different types and sizes are mixed together, move in different directions and often occlude each other in congested situations. This is, however, the reality of the urban traffic that has to be addressed. The promising ways to go are improvement of resolution of the video recordings, use of several cameras to get different views of the scene and cover larger area and fusion of video with other types of sensors (for example thermo-cameras or lidar). In a meantime, a combination of a simple but robust automated watchdog that can filter out situations that might be relevant for more detailed examination and manual or semi-automated post-processing of the filtered data seems to be a feasible alternative. (A)


  • English

Media Info

  • Media Type: Print
  • Features: Figures; References;
  • Pagination: pp 46-8
  • Monograph Title: Fit to drive. Proceedings of the 8th International traffic expert congress. Warsaw, May 8th-9th 2014
  • Serial:

Subject/Index Terms

Filing Info

  • Accession Number: 01597278
  • Record Type: Publication
  • Source Agency: Bundesanstalt für Straßenwesen (BASt)
  • ISBN: 978-3-2812-1918-2
  • Files: ITRD
  • Created Date: Mar 22 2016 3:17AM