next up previous
Next: Initialisation Up: Outline of the (Edge Previous: Outline of the (Edge

Detection of an Object Image Candidat (OIC)

  A short subsequence of image frames is used to estimate an OF field. This OF is segmented into regions. The basic idea consists in clustering OF vectors which are close and virtually parallel to each other. A cluster should comprise OF vectors which the difference between the starting points of neighboring OF vectors are closer to each other than a distance threshold and to which the angular difference between neighboring OF vectors does not exceed a small tolerance. Clusters which do not contain a minimum number of OF vectors will be rejected as being due to noise effects. Each accepted cluster will be enclosed by a rectangle. Rectangles which overlap each other are merged into an encompassing enclosing rectangle. The resulting rectangular region is considered to represent an OIC. The average direction of OF vectors aggregated into an OIC is taken to indicate the image plane velocity of the OIC.

  Whereas the two subprocesses outlined in the preceding paragraph operate in the Picture Domain (PD) (see, e. g., [Kanade 78]), the next subprocess in this initialisation process exploits knowledge about the camera as well as about its position and orientation relative to the recorded scene - i. e. about the internal and external camera parameters, respectively - to be obtained by a separate camera calibration step.

HHN (Karlsruhe): Conceptual Descriptions from Image Sequences (Dec. 16, 1996)