Video conference systems implementing orchestration models
A method for generating an output video stream in a video conference comprising receiving a plurality of input video streams of the video conference, receiving a series of observation events ( 52, 53, 54 ), the observation corresponding to actions made by participants of the video conference, Providing a plurality of orchestration models, Determining, for each of the orchestration models a probability of the series of observation events received, Selecting an orchestration model corresponding to the highest probability, Using the selected orchestration model to perform the steps of: • selecting the display state ( 51, 40, 41, 42 ) as a candidate display state, • Determining a conditional probability of the candidate display state for the received series of observation events • Determining the candidate display state providing the highest conditional probability as an updated display state, • Generating a video stream comprising the current display state and the updated display state.
1. A method for generating an output video stream in a video conference comprising:
Receiving a plurality of input video streams of the video conference
Receiving a series of observation events, the observation events belonging to a plurality of observable actions corresponding to actions made by participants of the video conference,
Providing a plurality of orchestration models, each model comprising:
A set of display states, each one associated with a predefined screen template, each screen template comprising a selected subset of the input video streams,
Transition probabilities between the display states,
Observation probabilities representing the conditional probabilities of the observable actions as a function of the display states,
Determining, for each of the orchestration models a probability of the series of observation events received,
Selecting an orchestration model corresponding to the highest probability
Using the selected orchestration model to perform:
For each display state of the orchestration model, selecting the display state as a candidate display state,
Determining a conditional probability of the candidate display state for the received series of observation events taking into account a sequence of display states including past display states and a current display state,
Determining the candidate display state providing the highest conditional probability as an updated display state,
Generating a video stream comprising one after the other a first sequence of images representing the screen template associated to the current display state and a second sequence of images representing the screen template associated to the updated display state.
2. A method according to claim 1 , wherein the observable actions are selected in the group of action categories of gestures, head motions, face expressions, audio actions, enunciation of keywords, actions relating to presentation slides.
3. A method according to claim 1 , wherein the observable actions are selected in the group of:
raising a finger, raising a hand,
making a head top down movement, making a head right left movement,
making a face expression that corresponds to speaking or sleeping,
making a noise, making silence, speaking by the tutor, speaking by a participant,
enunciating a name of an auditor or a subtitle,
switching a slide, moving a pointer,
beginning a question, ending a question.
4. A method in accordance with claim 1 , wherein the input video streams are selected in a group of: views of individual participants, views of a speaker, views of a conference room and views of presentation slides.
5. A method in accordance with claim 1 , wherein a screen template comprises a predefined arrangement of the input video streams belonging to the corresponding subset.
6. A method in accordance with claim 1 , wherein the transition probabilities are arranged as a transition matrix.
7. A method in accordance with claim 1 , wherein observation probabilities are arranged as an emission matrix.
8. A video conference control device for generating an output video stream in a video conference, the device comprising:
Means for receiving a plurality of input video streams of the video conference,
Means for receiving a series of observation events, the observation events belonging to a plurality of observable actions corresponding to actions made by participants of the video conference,
A data repository storing a plurality of orchestration models, each model comprising:
A set of display states, each one associated with a predefined screen template, each screen template comprising a selected subset of the input video streams,
Transition probabilities between the display states,
Observation probabilities representing the conditional probabilities of the observable actions as a function of the display states,
Means for determining, for each of the orchestration models, a probability of the series of observation events received,
Means for selecting an orchestration model corresponding to the highest probability,
Means for using the selected orchestration model to perform the steps of:
For each display state of the orchestration model, selecting the display state as a candidate display state,
Determining a conditional probability of the candidate display state for the received series of observation events taking into account a sequence of display states including past display states and a current display state,
Determining the candidate display state providing the highest conditional probability as an updated display state,
Generating a video stream comprising one after the other a first sequence of images representing the screen template associated to the current display state and a second sequence of images representing the screen template associated to the updated display state.
9. A video conference control device according to claim 8 , wherein the observable actions are selected in the group of action categories of gestures, head motions, face expressions, audio actions, enunciation of keywords, actions relating to presentation slides.
10. A video conference control device in accordance with claim 8 , wherein the observable actions are selected in the group of:
raising a finger, raising a hand,
making a head top down movement, making a head right left movement,
making a face expression that corresponds to speaking or sleeping,
making a noise, making silence, speaking by the tutor, speaking by a participant,
enunciating a name of an auditor or a subtitle,
switching a slide, moving a pointer,
beginning a question, ending a question.
11. A video conference control device in accordance with claim 8 , wherein the input video streams are selected in a group of: views of individual participants, views of a speaker, views of a conference room and views of presentation slides.
12. A video conference control device in accordance with claim 8 , wherein a screen template comprises a predefined arrangement of the input video streams belonging to the corresponding subset.
13. A video conference control device in accordance with claim 8 , wherein the transition probabilities are arranged as a transition matrix.
14. A video conference control device in accordance with claim 8 , wherein observation probabilities are arranged as an emission matrix.
15. A video conference system comprising a video conference control device in accordance with claim 8 , connected by a communication network to a plurality of terminals, wherein each terminal comprises means for generating an input video stream and wherein the communication network is adapted to transmit the video stream from the terminals to the control device and to transmit the output video stream generated by the control device to a terminal.