Interactive visualization or interactive visualisation is a branch of graphic visualization in computer science that involves studying how humans interact with computers to create graphic illustrations of information and how this process can be made more efficient. For a visualization to be considered interactive it must satisfy two criteria:
Human input: control of some aspect of the visual representation of information, or of the information being represented, must be available to a human, and
Response time: changes made by the human must be incorporated into the visualization in a timely manner. In general, interactive visualization is considered a soft real-time task.
One particular type of interactive visualization is virtual reality, where the visual representation of information is presented using an immersive display device such as a stereo projector. VR is also characterized by the use of a spatial metaphor, where some aspect of the information is represented in three dimensions so that humans can explore the information as if it were present, sized appropriately, or had shape. Another type of interactive visualization is collaborative visualization, in which multiple people interact with the same computer visualization to communicate their ideas to each other or to explore information cooperatively. Frequently, collaborative visualization is used when people are physically separated. Using several networked computers, the same visualization can be presented to each person simultaneously. The people then make annotations to the visualization as well as communicate via audio, video, or text messages.
All of these actions require a physical device. Input devices range from the common – keyboards, mice, graphics tablets, trackballs, and touchpads – to the esoteric – wired gloves, boom arms, and even omnidirectional treadmills. These input actions can be used to control both the information being represented or the way that the information is presented. When the information being presented is altered, the visualization is usually part of a feedback loop. For example, consider an aircraft avionics system where the pilot inputs roll, pitch, and yaw and the visualization system provides a rendering of the aircraft's new attitude. Another example would be a scientist who changes a simulation while it is running in response to a visualization of its current progress. This is called computational steering. More frequently, the representation of the information is changed rather than the information itself.
Rapid response to human input
Experiments have shown that a delay of more than 20 ms between when input is provided and a visual representation is updated is noticeable by most people. Thus it is desirable for an interactive visualization to provide a rendering based on human input within this time frame. However, when large amounts of data must be processed to create a visualization, this becomes hard or even impossible with current technology. Thus the term "interactive visualization" is usually applied to systems that provide feedback to users within several seconds of input. The term interactive framerate is often used to measure how interactive a visualization is. Framerates measure the frequency with which an image can be generated by a visualization system. A framerate of 50 frames per second is considered good while 0.1 frame/s would be considered poor. The use of framerates to characterize interactivity is slightly misleading however, since framerate is a measure of bandwidth while humans are more sensitive to latency. Specifically, it is possible to achieve a good framerate of 50 frame/s but if the images generated refer to changes to the visualization that a person made more than 1 second ago, it will not feel interactive to a person. The rapid response time required for interactive visualization is a difficult constraint to meet and there are several approaches that have been explored to provide people with rapid visual feedback based on their input. Some include
Parallel rendering – where more than one computer or video card is used simultaneously to render an image. Multiple frames can be rendered at the same time by different computers and the results transferred over the network for display on a single monitor. This requires each computer to hold a copy of all the information to be rendered and increases bandwidth, but also increases latency. Also, each computer can render a different region of a single frame and send the results over a network for display. This again requires each computer to hold all of the data and can lead to a load imbalance when one computer is responsible for rendering a region of the screen with more information than other computers. Finally, each computer can render an entire frame containing a subset of the information. The resulting images plus the associated depth buffer can then be sent across the network and merged with the images from other computers. The result is a single frame containing all the information to be rendered, even though no single computer's memory held all of the information. This is called parallel depth compositing and is used when large amounts of information must be rendered interactively.
Progressive rendering – where a framerate is guaranteed by rendering some subset of the information to be presented and providing incremental improvements to the rendering once the visualization is no longer changing.
Level-of-detail rendering – where simplified representations of information are rendered to achieve a desired framerate while a person is providing input and then the full representation is used to generate a still image once the person is through manipulating the visualization. One common variant of LOD rendering is subsampling. When the information being represented is stored in a topologically rectangular array, a lower resolution version can easily be generated by skipping n points for each 1 point rendered. Subsampling can also be used to accelerate rendering techniques such as volume visualization that require more than twice the computations for an image twice the size. By rendering a smaller image and then scaling the image to fill the requested screen space, much less time is required to render the same data.
Frameless rendering – where the visualization is no longer presented as a time series of images, but as a single image where different regions are updated over time.
