Inventor(s)

Abstract

A client playing streamed audiovisual content estimates whether a viewer is attending to a display by using a short-range radar on the same device as the display, or on a companion device in the same scene. The radar is of a millimetre-wave, frequency-modulated continuous-wave, or ultra-wideband type. No camera is used, and no image of the scene is formed or sent. The attention proxy is presence in the scene together with an estimate of whether the head or torso is oriented toward the display. The proxy is not a fine measurement of eye gaze. When the proxy, after a dwell and hysteresis interval of several seconds, indicates that the viewer is not attending, the client steps the requested adaptive-bitrate rendition or bitrate down, and may dim the backlight, later turn the backlight off, continue with audio only, or pause or stop playback. Stages distinguish a head turned away or looking at another device, a viewer who has left the room, and a viewer who is still present but still or asleep, the last inferred from respiration or other micro-motion. Optional secondary signals include treating a handheld object as a strong radar reflector and receiving a content-free “actively in use” flag from another device on the home network. An optional semantic path may hold a low-rate description of the programme and reconstruct the picture locally. In that path, the attention proxy also sets the receiver's generation cost: full-quality reconstruction when the viewer is attending; a more energy-efficient method — lower output resolution or frame rate, a smaller model, fewer inference steps, or keyframe-only or reduced-detail rendering — when attention is only partial, for example while looking at a phone; and no video generation, with audio only and the backlight off, when the viewer has left the room but is still listening, the semantic stream then optionally being reduced to a state that supports a fast resume. When attention returns, the default restore is to prefetch a higher-definition rendition and switch to it at the next segment boundary. The already-buffered low-rung media are retained as a fallback, so that if the viewer looks away again playback can drop back to that low rendition at a segment boundary; higher-rung segments are fetched just-in-time for the expected attended period. On-device super-resolution or upscaling of those buffered low-rung segments is an optional bridge only, not the default restore. An optional predictive path may use aggregated, anonymised, content-free attention statistics, and a per-viewer history, to prefetch or pre-switch a higher rendition ahead of segments that typically draw attention, and to keep a lower rendition during typically low-attention segments. Optionally, display or backlight brightness may track estimated downward head pitch continuously, lowering as the head tilts down and returning as it tilts back up, with smoothing or a rate limit and a minimum floor; if the display is the only or dominant light source in the room, that reduction is not applied, or is limited to a safe floor, so the room is not plunged into darkness. Illustrative arithmetic examples of bandwidth in gigabytes and of site energy at the meter in kilowatt-hours are given; they are not measurements.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.

Share

COinS