next up previous
Next: Outline of the (Edge Up: No Title Previous: Introductory Remarks


Overview of a Systems Approach Towards

We shall use the DA of road traffic in order to illustrate the decomposition of the algorithmic process EIS into subprocesses. Each subprocess has to solve a particular part of the overall problem.

  1. Detection: How to associate a subregion within an image frame with the hypothesis that this subregion represents the image of a vehicle   - an object image candidate (OIC) - which moves in the recorded scene with respect to the stationary camera ?
  2. Object description: How to represent the shape of each vehicle type which will be admitted in the scene within the field of view ?
  3. Description of the environment: How to represent the geometry of the road environment within which the vehicles move while the camera records the road scene ?
  4. Classification: Association of an OIC with one of the admitted vehicle types.
  5. Initialisation: How to use an OIC in order to generate initial estimates for the position, orientation, size, shape, and velocity of a 3D vehicle model in the scene domain ?
  6. Tracking: How to track the selected vehicle by model-based computer vision ?
  7. Occurrence description: How to associate segments of a vehicle trajectory with motion verbs ?
  8. Recognition of temporal developments: How to concatenate selected occurrences in such a manner that a coherent description of a sequence of vehicle manouvers can be generated ?
  9. Recognition of situations: How can an algorithm decide whether an agent - i. e. a vehicle in our DA - will continue its current manouver or switch to a different one ?
  10. Recognition of the goal of an agent: Can future subsequences of manouvers of an agent (i. e. of a vehicle in our DA) be inferred on the basis of already observed manouver subsequences ?
  11. Generation of a natural language text describing the temporal development of activities of each agent as well as of various aggregations of agents, i. e. groups of vehicles, at different levels of abstraction.
For the current initial version of this tutorial, the discussion will concentrate on the detection, initialisation, and tracking of vehicles. The description of road environments, the classification of object (image) candidates as well as the transition from geometric descriptions towards natural language text will be treated as soon as possible. In order to get this tutorial off the ground, certain topics such as change detection will initially just be hinted at or skipped entirely. In this manner, it is hoped to sketch an overall systems approach before details are filled in subsequently.

next up previous
Next: Outline of the (Edge Up: No Title Previous: Introductory Remarks

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