Dokument-ID Dokumenttyp Verfasser/Autoren Herausgeber Haupttitel Abstract Auflage Verlagsort Verlag Erscheinungsjahr Seitenzahl Schriftenreihe Titel Schriftenreihe Bandzahl ISBN Quelle der Hochschulschrift Konferenzname Quelle:Titel Quelle:Jahrgang Quelle:Heftnummer Quelle:Erste Seite Quelle:Letzte Seite URN DOI Abteilungen OPUS4-2978 Konferenzveröffentlichung Knauer, Uwe; Dammeier, T.; Meffert, Beate Gürlebeck, Klaus; Könke, Carsten THE STRUCTURE OF ROAD TRAFFIC SCENES AS REVEALED BY UNSUPERVISED ANALYSIS OF THE TIME AVERAGED OPTICAL FLOW The Lucas-Kanade tracker has proven to be an efficient and accurate method for calculation of the optical flow. However, this algorithm can reliably track only suitable image features like corners and edges. Therefore, the optical flow can only be calculated for a few points in each image, resulting in sparse optical flow fields. Accumulation of these vectors over time is a suitable method to retrieve a dense motion vector field. However, the accumulation process limits application of the proposed method to fixed camera setups. Here, a histogram based approach is favored to allow more than a single typical flow vector per pixel. The resulting vector field can be used to detect roads and prescribed driving directions which constrain object movements. The motion structure can be modeled as a graph. The nodes represent entry and exit points for road users as well as crossings, while the edges represent typical paths. 9 urn:nbn:de:gbv:wim2-20170327-29780 10.25643/bauhaus-universitaet.2978 In Zusammenarbeit mit der Bauhaus-Universität Weimar