The development of a driver assistant system supporting drivers in complex intersection situations would be a major achievement for traffic safety. This thesis focused on the visual detection of independently moving objects, for two reasons: (1) Humans gather their information to a large extent visually, (2) cameras are inexpensive and often used in vehicles. So a small lightweight monocular camera system has been chosen as input device. Information about moving objects has been used to develop a prototype of an attention guidance system, which is a central component of any intersection assistant system. It is based on the analysis of sequences from a single freely moving camera, rigidly coupled with measurements from inertial sensors. The system comprises three major parts, which are discussed in detail: the estimation of egomotion of the camera relative to the static scene; the detection of independent motion based on image point correspondences; a Bayesian framework to integrate the measurements in a consistent way. Both, the egomotion estimation methods and the detection algorithms for independent motion operate on correspondences between points in images. An algorithm for the detection and tracking of such feature points has been chosen. These correspondences are computed using an algorithm which iteratively minimises the intensity differences in a small support window (Kanade-Lucas-Tomasi feature tracking). Correspondence computation is only unambiguous if enough structure is present in the support window, and hence a corner detector is used to identify promising regions for correspondence estimation. Even though both algorithms (i.e. egomotion and detection) are independent of the particular method used for feature point tracking, the chosen tracking algorithm must be able to deal with the aperture problem and must result in full correspondences. A prototype was build with the detection system extended by a simple spatial clustering of points belonging to a single object. Objects are tracked over time using a Kalman filter, and finally a warning is triggered when an object an a collision course is detected. The prototype of the attention guidance system has been integrated into the UTA (Urban Traffic Assistant) demonstrator from the DaimlerChrysler AG. The prototype has been presented in over 200 successful live runs to representatives from politics and industry including the German minister of education and science.


    Zugriff

    Zugriff über TIB

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Visual detection of independently moving objects by a moving monocular observer


    Weitere Titelangaben:

    Visuelle Erkennung sich unabhängig bewegender Objekte durch einen sich bewegenden monokularen Beobachter


    Beteiligte:

    Erschienen in:

    Erscheinungsdatum :

    2008


    Format / Umfang :

    188 Seiten, Bilder, Tabellen, Quellen


    Anmerkungen:

    S. Anhang)




    Medientyp :

    Hochschulschrift


    Format :

    Print


    Sprache :

    Englisch






    Detection of Independently Moving Objects in Passive Video

    Morimoto, C. / De Menthon, D. / Davis, L. et al. | British Library Conference Proceedings | 1995


    A Multibody Factorization Method for Independently Moving Objects

    Costeira, J. P. / Kanade, T. | British Library Online Contents | 1998


    Moving Obstacle Detection using Monocular Vision

    Yamaguchi, K. / Kato, T. / Ninomiya, Y. | IEEE | 2006