Welcome to the project page of

Automatic Lecture Recording

Some more details on the director module:

The procedure to determine the next transition to the next state is done by the following steps:

  1. Determination of all possible transitions going out of the active state.
  2. Initialize the probabilistic values of these transitions.
  3. Decrease the values dependent on the history of the last active states and their comparison with the next state of the certain transition. Less decrease for a state that has be shown less recently.
  4. Compare the condition objects and conditions with the sensorinputs and manipulate the according values. A sensor reports back the condition object, the condition and a percentage. For example, the indoor positioning system detects a walking lecturer.
  5. Manipulate the values depending on the condition objects and conditions compared to the cameraman's feedback. For example, the lecturer cameraman detects a motion in 20% of the whole image which can be interpreted as a gesticulating lecturer.
  6. Change the values depending on the cameras' status reports. In case a camera is still moving to its next target, it feedbacks its motion so that this transition becomes less likely, as long as the camera is moving.
  7. Check for hard events, e.g., the lecturer gives the floor to a questioner. All transitions leading to the question context get their values increased, especially if they are going to show the questioner.
  8. Determine all transitions with the highest probabilistic value. If there are more than one, choose one of them randomly.

In this procedure the director behaves always similar but seldom identical and the next transition becomes less predictive.