Meeting Chronicler

SRI has been developing a Meeting Chronicler, that records the audio and video of meetings, then automatically summarizes and indexes their contents for later search and retrieval. The Meeting Chronicler will help a viewer to quickly comprehend the important aspects of a meeting and review particular segments of interest. The chronicler could be used in classroom, courtroom, office, and video conference environments.

Before a meeting, multiple cameras and microphones are placed in the meeting room to capture the audio and video. During the meeting, individual participants are tracked, and their activities are detected, monitored, and recorded. One or more pan-tilt-zoom cameras automatically follow and capture close-ups of the participants and their actions. After the meeting, automatic speech recognition and natural language processing software is run on the audio data to produce a transcript of the meeting and segment it into topics.

The figure shows a prototype interface for accessing the meeting information. The meeting is indexed by activities and topics, which allows video and audio segments associated with these items to be retrieved and replayed.

Meeting Chronicler Prototype Interface [click to view enlarged image]

Meeting Chronicler is based on SRI technologies for people tracking, video event detection, speech recognition, and natural language processing. The video processing occurs in real time on a single PC. Inputs from three stereo video cameras are processed to compute the 3-D positions of many points in a scene. The 3-D data is used to detect, track, and localize people moving about in the room. The known positions and poses of the people relative to places in the room are used to detect visual events relevant to chronicling meetings, such as "person 1 sat down at the table," "person 3 left the room," or "person 2 went to the whiteboard." In the video, the left side shows views from the three cameras mounted about room. The middle part shows the layout of the room and the positions of people relative to room features
(a cross (+) over a person means that he or she is sitting down). The right side of the video shows the visual events that are automatically monitored and logged by the system.

• Video of real-time people tracking and event detection
  [Meeting Chronicler]