Use Cases for Telepresence Multistreams

The use cases focus on typical implementations. There are a number of possible variants for these use cases; for example, the audio supported may differ at the end points (such as mono or stereo versus surround sound), etc. Many of these systems offer a "full conference room" solution, where local participants sit at one side of a table and remote participants are displayed as if they are sitting on the other side of the table. The cameras and screens are typically arranged to provide a panoramic view of the remote room (left to right from the local user's viewpoint). The sense of immersion and non-verbal communication is fostered by a number of technical features, such as: Good eye contact, which is achieved by careful placement of participants, cameras, and screens. Camera field of view and screen sizes are matched so that the images of the remote room appear to be full size. The left side of each room is presented on the right screen at the far end; similarly, the right side of the room is presented on the left screen. The effect of this is that participants of each site appear to be sitting across the table from each other. If 2 participants on the same site glance at each other, all participants can observe it. Likewise, if a participant at one site gestures to a participant on the other site, all participants observe the gesture itself and the participants it includes.

In this case, each of the 2 sites has an identical number of screens, with cameras having fixed fields of view, and 1 camera for each screen. The sound type is the same at each end. As an example, there could be 3 cameras and 3 screens in each room, with stereo sound being sent and received at each end. Each screen is paired with a corresponding camera. Each camera/screen pair is typically connected to a separate codec, producing an encoded stream of video for transmission to the remote site, and receiving a similarly encoded stream from the remote site. Each system has one or multiple microphones for capturing audio. In some cases, stereophonic microphones are employed. In other systems, a microphone may be placed in front of each participant (or pair of participants). In typical systems, all the microphones are connected to a single codec that sends and receives the audio streams as either stereo or surround sound. The number of microphones and the number of audio channels are often not the same as the number of cameras. Also, the number of microphones is often not the same as the number of loudspeakers. The audio may be transmitted as multi-channel (stereo/surround sound) or as distinct and separate monophonic streams. Audio levels should be matched, so the sound levels at both sites are identical. Loudspeaker and microphone placements are chosen so that the sound "stage" (orientation of apparent audio sources) is coordinated with the video. That is, if a participant at one site speaks, the participants at the remote site perceive her voice as originating from her visual image. In order to accomplish this, the audio needs to be mapped at the received site in the same fashion as the video. That is, audio received from the right side of the room needs to be output from loudspeaker(s) on the left side at the remote site, and vice versa.

In this case, each site has a different number of screens and cameras than the other site. The important characteristic of this scenario is that the number of screens is different between the 2 sites. This creates challenges that are handled differently by different telepresence systems. This use case builds on the basic scenario of 3 screens to 3 screens. Here, we use the common case of 3 screens and 3 cameras at one site, and 1 screen and 1 camera at the other site, connected by a point-to-point call. The screen sizes and camera fields of view at both sites are basically similar, such that each camera view is designed to show 2 people sitting side by side. Thus, the 1-screen room has up to 2 people seated at the table, while the 3-screen room may have up to 6 people at the table. The basic considerations of defining left and right and indicating relative placement of the multiple audio and video streams are the same as in the 3-3 use case. However, handling the mismatch between the 2 sites of the number of screens and cameras requires more complicated maneuvers. For the video sent from the 1-camera room to the 3-screen room, usually what is done is to simply use 1 of the 3 screens and keep the second and third screens inactive or, for example, put up the current date. This would maintain the "full-size" image of the remote side. For the other direction, the 3-camera room sending video to the 1-screen room, there are more complicated variations to consider. Here are several possible ways in which the video streams can be handled. The 1-screen system might simply show only 1 of the 3 camera images, since the receiving side has only 1 screen. 2 people are seen at full size, but 4 people are not seen at all. The choice of which one of the 3 streams to display could be fixed, or could be selected by the users. It could also be made automatically based on who is speaking in the 3-screen room, such that the people in the 1-screen room always see the person who is speaking. If the automatic selection is done at the sender, the transmission of streams that are not displayed could be suppressed, which would avoid wasting bandwidth. The 1-screen system might be capable of receiving and decoding all 3 streams from all 3 cameras. The 1-screen system could then compose the 3 streams into 1 local image for display on the single screen. All 6 people would be seen, but smaller than full size. This could be done in conjunction with reducing the image resolution of the streams, such that encode/decode resources and bandwidth are not wasted on streams that will be downsized for display anyway. The 3-screen system might be capable of including all 6 people in a single stream to send to the 1-screen system. For example, it could use PTZ (Pan Tilt Zoom) cameras to physically adjust the cameras such that 1 camera captures the whole room of 6 people. Or, it could recompose the 3 camera images into 1 encoded stream to send to the remote site. These variations also show all 6 people but at a reduced size. Or, there could be a combination of these approaches, such as simultaneously showing the speaker in full size with a composite of all 6 participants in a smaller size. The receiving telepresence system needs to have information about the content of the streams it receives to make any of these decisions. If the systems are capable of supporting more than one strategy, there needs to be some negotiation between the 2 sites to figure out which of the possible variations they will use in a specific point-to-point call.

In a multipoint telepresence conference, there are more than 2 sites participating. Additional complexity is required to enable media streams from each participant to show up on the screens of the other participants. Clearly, there are a great number of topologies that can be used to display the streams from multiple sites participating in a conference. One major objective for telepresence is to be able to preserve the "being there" user experience. However, in multi-site conferences, it is often (in fact, usually) not possible to simultaneously provide full-size video, eye contact, and common perception of gestures and gaze by all participants. Several policies can be used for stream distribution and display: all provide good results, but they all make different compromises. One common policy is called site switching. Let's say the speaker is at site A and the other participants are at various "remote" sites. When the room at site A shown, all the camera images from site A are forwarded to the remote sites. Therefore, at each receiving remote site, all the screens display camera images from site A. This can be used to preserve full-size image display, and also provide full visual context of the displayed far end, site A. In site switching, there is a fixed relation between the cameras in each room and the screens in remote rooms. The room or participants being shown are switched from time to time based on who is speaking or by manual control, e.g., from site A to site B. Segment switching is another policy choice. In segment switching (assuming still that site A is where the speaker is, and "remote" refers to all the other sites), rather than sending all the images from site A, only the speaker at site A is shown. The camera images of the current speaker and previous speakers (if any) are forwarded to the other sites in the conference. Therefore, the screens in each site are usually displaying images from different remote sites -- the current speaker at site A and the previous ones. This strategy can be used to preserve full-size image display and also capture the non-verbal communication between the speakers. In segment switching, the display depends on the activity in the remote rooms (generally, but not necessarily based on audio/speech detection). A third possibility is to reduce the image size so that multiple camera views can be composited onto one or more screens. This does not preserve full-size image display, but it provides the most visual context (since more sites or segments can be seen). Typically in this case, the display mapping is static, i.e., each part of each room is shown in the same location on the display screens throughout the conference. Other policies and combinations are also possible. For example, there can be a static display of all screens from all remote rooms, with part or all of one screen being used to show the current speaker at full size.

In addition to the video and audio streams showing the participants, additional streams are used for presentations. In systems available today, generally only one additional video stream is available for presentations. Often, this presentation stream is half-duplex in nature, with presenters taking turns. The presentation stream may be captured from a PC screen, or it may come from a multimedia source such as a document camera, camcorder, or a DVD. In a multipoint meeting, the presentation streams for the currently active presentation are always distributed to all sites in the meeting, so that the presentations are viewed by all. Some systems display the presentation streams on a screen that is mounted either above or below the 3 participant screens. Other systems provide screens on the conference table for observing presentations. If multiple presentation screens are used, they generally display identical content. There is considerable variation in the placement, number, and size of presentation screens. In some systems, presentation audio is pre-mixed with the room audio. In others, a separate presentation audio stream is provided (if the presentation includes audio). In H.323 systems, H.239 is typically used to control the video presentation stream. In SIP systems, similar control mechanisms can be provided using the Binary Floor Control Protocol (BFCP) for the presentation token. These mechanisms are suitable for managing a single presentation stream. Although today's systems remain limited to a single video presentation stream, there are obvious uses for multiple presentation streams: Frequently, the meeting convener is following a meeting agenda, and it is useful for her to be able to show that agenda to all participants during the meeting. Other participants at various remote sites are able to make presentations during the meeting, with the presenters taking turns. The presentations and the agenda are both shown, either on separate screens, or perhaps rescaled and shown on a single screen. A single multimedia presentation can itself include multiple video streams that should be shown together. For instance, a presenter may be discussing the fairness of media coverage. In addition to slides that support the presenter's conclusions, she also has video excerpts from various news programs that she shows to illustrate her findings. She uses a DVD player for the video excerpts so that she can pause and reposition the video as needed. An educator who is presenting a multiscreen slide show. This show requires that the placement of the images on the multiple screens at each site be consistent. There are many other examples where multiple presentation streams are useful.

It is common in meeting scenarios for people to join the conference from a variety of environments, using different types of endpoint devices. A multiscreen immersive telepresence conference may include someone on a PC-based video conferencing system, a participant calling in by phone, and (soon) someone on a handheld device. What experience/view will each of these devices have? Some may be able to handle multiple streams, and others can handle only a single stream. (Here, we are not talking about legacy systems, but rather systems built to participate in such a conference, although they are single stream only.) In a single video stream, the stream may contain one or more compositions depending on the available screen space on the device. In most cases, an intermediate transcoding device will be relied upon to produce a single stream, perhaps with some kind of continuous presence. Bit rates will vary -- the handheld device and phone having lower bit rates than PC and multiscreen systems. Layout is accomplished according to different policies. For example, a handheld device and PC may receive the active speaker stream. The decision can either be made explicitly by the receiver or by the sender if it can receive some kind of rendering hint. The same is true for audio -- i.e., that it receives a mixed stream or a number of the loudest speakers if mixing is not available in the network. For the PC-based conferencing participant, the user's experience depends on the application. It could be single stream, similar to a handheld device but with a bigger screen. Or, it could be multiple streams, similar to an immersive telepresence system but with a smaller screen. Control for manipulation of streams can be local in the software application, or in another location and sent to the application over the network. The handheld device is the most extreme. How will that participant be viewed and heard? It should be an equal participant, though the bandwidth will be significantly less than an immersive system. A receiver may choose to display output coming from a handheld device differently based on the resolution, but that would be the case with any low-resolution video stream, e.g., from a powerful PC on a bad network. The handheld device will send and receive a single video stream, which could be a composite or a subset of the conference. The handheld device could say what it wants or could accept whatever the sender (conference server or sending endpoint) thinks is best. The handheld device will have to signal any actions it wants to take the same way that an immersive system signals actions.

The importance of this example is that the multiple video streams are not used to create an immersive conferencing experience with panoramic views at all the sites. Instead, the multiple streams are dynamically used to enable full participation of remote students in a university class. In some instances, the same video stream is displayed on multiple screens in the room; in other instances, an available stream is not displayed at all. The main site is a university auditorium that is equipped with 3 cameras. One camera is focused on the professor at the podium. A second camera is mounted on the wall behind the professor and captures the class in its entirety. The third camera is co-located with the second and is designed to capture a close-up view of a questioner in the audience. It automatically zooms in on that student using sound localization. Although the auditorium is equipped with 3 cameras, it is only equipped with 2 screens. One is a large screen located at the front so that the class can see it. The other is located at the rear so the professor can see it. When someone asks a question, the front screen shows the questioner. Otherwise, it shows the professor (ensuring everyone can easily see her). The remote sites are typical immersive telepresence rooms, each with 3 camera/screen pairs. All remote sites display the professor on the center screen at full size. A second screen shows the entire classroom view when the professor is speaking. However, when a student asks a question, the second screen shows the close-up view of the student at full size. Sometimes the student is in the auditorium; sometimes the speaking student is at another remote site. The remote systems never display the students that are actually in that room. If someone at a remote site asks a question, then the screen in the auditorium will show the remote student at full size (as if they were present in the auditorium itself). The screen in the rear also shows this questioner, allowing the professor to see and respond to the student without needing to turn her back on the main class. When no one is asking a question, the screen in the rear briefly shows a full-room view of each remote site in turn, allowing the professor to monitor the entire class (remote and local students). The professor can also use a control on the podium to see a particular site -- she can choose either a full-room view or a single-camera view. Realization of this use case does not require any negotiation between the participating sites. Endpoint devices (and a Multipoint Control Unit (MCU), if present) need to know who is speaking and what video stream includes the view of that speaker. The remote systems need some knowledge of which stream should be placed in the center. The ability of the professor to see specific sites (or for the system to show all the sites in turn) would also require the auditorium system to know what sites are available and to be able to request a particular view of any site. Bandwidth is optimized if video that is not being shown at a particular site is not distributed to that site.

This use case describes a virtual space multipoint meeting with good eye contact and spatial layout of participants. The use case was proposed very early in the development of video conferencing systems as described in 1983 by Allardyce and Randal . The use case is illustrated in Figure 2-5 of their report. The virtual space expands the point-to-point case by having all multipoint conference participants "seated" in a virtual room. In this case, each participant has a fixed "seat" in the virtual room, so each participant expects to see a different view having a different participant on his left and right side. Today, the use case is implemented in multiple telepresence-type video conferencing systems on the market. The term "virtual space" was used in their report. The main difference between the result obtained with modern systems and those from 1983 are larger screen sizes. Virtual space multipoint as defined here assumes endpoints with multiple cameras and screens. Usually, there is the same number of cameras and screens at a given endpoint. A camera is positioned above each screen. A key aspect of virtual space multipoint is the details of how the cameras are aimed. The cameras are each aimed on the same area of view of the participants at the site. Thus, each camera takes a picture of the same set of people but from a different angle. Each endpoint sender in the virtual space multipoint meeting therefore offers a choice of video streams to remote receivers, each stream representing a different viewpoint. For example, a camera positioned above a screen to a participant's left may take video pictures of the participant's left ear; while at the same time, a camera positioned above a screen to the participant's right may take video pictures of the participant's right ear. Since a sending endpoint has a camera associated with each screen, an association is made between the receiving stream output on a particular screen and the corresponding sending stream from the camera associated with that screen. These associations are repeated for each screen/camera pair in a meeting. The result of this system is a horizontal arrangement of video images from remote sites, one per screen. The image from each screen is paired with the camera output from the camera above that screen, resulting in excellent eye contact.

This use case describes a scenario where multiple presentation streams are used. In this use case, the local site is a surgery room connected to one or more remote sites that may have different capabilities. At the local site, 3 main cameras capture the whole room (the typical 3-camera telepresence case). Also, multiple presentation inputs are available: a surgery camera that is used to provide a zoomed view of the operation, an endoscopic monitor, a flouroscope (X-ray imaging), an ultrasound diagnostic device, an electrocardiogram (ECG) monitor, etc. These devices are used to provide multiple local video presentation streams to help the surgeon monitor the status of the patient and assist in the surgical process. The local site may have 3 main screens and one (or more) presentation screen(s). The main screens can be used to display the remote experts. The presentation screen(s) can be used to display multiple presentation streams from local and remote sites simultaneously. The 3 main cameras capture different parts of the surgery room. The surgeon can decide the number, the size, and the placement of the presentations displayed on the local presentation screen(s). He can also indicate which local presentation captures are provided for the remote sites. The local site can send multiple presentation captures to remote sites, and it can receive from them multiple presentations related to the patient or the procedure. One type of remote site is a single- or dual-screen and one-camera system used by a consulting expert. In the general case, the remote sites can be part of a multipoint telepresence conference. The presentation screens at the remote sites allow the experts to see the details of the operation and related data. Like the main site, the experts can decide the number, the size, and the placement of the presentations displayed on the presentation screens. The presentation screens can display presentation streams from the surgery room, from other remote sites, or from local presentation streams. Thus, the experts can also start sending presentation streams that can carry medical records, pathology data, or their references and analysis, etc. Another type of remote site is a typical immersive telepresence room with 3 camera/screen pairs, allowing more experts to join the consultation. These sites can also be used for education. The teacher, who is not necessarily the surgeon, and the students are in different remote sites. Students can observe and learn the details of the whole procedure, while the teacher can explain and answer questions during the operation. All remote education sites can display the surgery room. Another option is to display the surgery room on the center screen, and the rest of the screens can show the teacher and the student who is asking a question. For all the above sites, multiple presentation screens can be used to enhance visibility: one screen for the zoomed surgery stream and the others for medical image streams, such as MRI images, cardiograms, ultrasonic images, and pathology data.