Die Eignung spezifischer Bildstrukturen für die Bewegungsbestimmung in Bildfolgen.

An algorithm for the detection of moving objects in a natural environment is described. It is based on the extraction of features in consecutive frames. Tracking feature positions in several frames yields a motion vector field that is analyzed for discontinuities indicating a moving object. Four feature detectors are investigated. Two ot them mark points of high variance and high gray value curvature. One detects points of high contrast and merges them to straight lines. The last one, developed especially for low contrast regions, depends only on the monotonicity of gray values but not on the value itself. The performance of each feature extractor is evaluated based on the detection results. Experiments with a sequence of 120 pictures of a moving car scene show that the combination of the straight line detector with the monotonicity operator performs better than each single operator. (IITB)