A reconstructed car is shown superimposed on one of the original images from which the conceptual description was prepared.

Generation of Synthetic Image Sequences for Car Maneouvers Extracted by Image Sequence Evaluation

V. Jeyakumar
April, 1999
Advisor:

Amitabha Mukerjee, IIT Kanpur, India and
Hans-Hellmut Nagel, Uni-Karlsruhe, Germany

 

Abstract

Systems that analyze video data generate voluminous output that is impossible to scrutinize manually. This work builds on past work that generates conceptual descriptions (abstractions) of visual events in the traffic domain. Based on this, synthetic image sequences are reconstructed for the car maneuvers. The parameterized vehicle model is projected onto a specific lane in a static image and is maneuvered by means of a motion model whose inputs are provided by decoding the conceptual description. Various cases from simple - i.e. a car on a straight road - to more complex maneuver sequences, for example lane changing, are to be analyzed. This provides an easy method for humans to verify the output of a video analysis. It is also of interest on its own as a reconstruction from abstract motion description, eg: in video search or in movie generation.

     H.-H. Nagel, M. Haag, V. Jeyakumar, Amitabha Mukerjee, 1999
     Visualization of conceptual descriptions derived from image sequences
     DAGR '99 (German Conf on Artificial Intelligence), Bonn,
          Germany, 13-15 Sept 1999