A video computing algorithm for stationary vehicle detection in tunnels
Abstract
To detect stationary vehicles is by far the most important application in traffic surveillance. Especially for tunnel surveillance, this task is crucial. This talk presents a pixel-wise method for stationary vehicle detection, which was developed for ARC Seibersdorf research. It is a combination of background modelling/maintenance and frame differencing. Stationary vehicles are pixel regions, that are foreground with respect to the background model and show significant intensity changes in the near past (vehicles move before they stop). The background model is adapted by certain constraints based on the actual video stream. The method is made robust by morphological voting for each pixel. The novelty of this approach is that the result of voting is used to correct the pixel's background model. Furthermore, the implications between latency of detection, quality of images (noise) and frame-rate are well defined by the parameters of the algorithm. In the end of the talk, qualitative results on tests videos of a truck stop and an accident in Kaisermuehlen tunnel are shown.
