G - Physics – 06 – T
Patent
G - Physics
06
T
G06T 7/00 (2006.01) G06T 1/00 (2006.01) H04N 3/15 (2006.01) H04N 5/217 (2006.01)
Patent
CA 2349828
A method, apparatus, and system for extracting a high-resolution, maximally detector-noise-removed image, along with texture, motion and velocity information, all directly from a detector array and its supporting logic, is described. The texture isolation, edge-motion detection, and noise removal processes rely largely upon principles deduced from mammalian vision. Spatially coincident opponent center/surround structures fed by the primary detector array are employed to remove spatio-temporally random detector-noise events. Static edges are accurately located by spatially oscillating the image with respect to the detector, filtering only for those edges whose motions reflect purely the induced oscillation, and obtaining accurate phase information at edge-crossings. Spatio-temporal activity in the image itself, not suitably matching the space-time characteristics of this oscillation (or a Doppler shifted version of it) can also be treated as noise if desired. During static periods (i.e. between induced oscillations), objects undergoing real motion detectable in the plane of the array can be selectively given attention by temporarily suppressing (either actively, or passively) those detectors having detected stationary edges during the previous oscillation, greatly reducing the computational load on robotic vision systems. Using Doppler techniques, accurate object velocities may be collected during oscillation, by detecting the frequency shift (with respect to the induced oscillation), of objects crossing multiple opponent center/surround detectors, and noting the orientation of the sequence of crossings. For a particular orientation of oscillation, center/surround crossings by a slice of texture in that orientation will yield a specific frequency spectrum of edge-crossing bursts. Given more than one orientation (and possibly scale) of such oscillation, and grouping regions by their resulting oscillation spectra, texture regions may be efficiently isolated and identified. Textures, edges and isolated points would have substantially different spectra, useful for identification. Oscillation of scale (zooming in and out) would permit radial velocity detection for simple collision avoidance detectors. A spatially- varying detection threshold, based upon localized image content, can be employed to suppress texture, to control entropy, or to match human vision characteristics. This detector does not employ raster scanning, but is instead event-driven, reporting exact timings and orientations of edge crossings from one opponent center/surround to the next; therefore, its spatial- and velocity-reporting accuracy are functions of its opponent center/surround temporal sensitivity, leaving open the possibility of using cheaper, larger pixels, with fewer pixels per detector array. Since actual velocities are reported in real-time from the array support logic, there is no need to calculate processor-intensive, inaccurate motion-vector fields (requiring a minimum of one frame of latency). Another benefit is the self-calibration of all adjacent detectors by systematic detector- difference cancellation during oscillation of the array, referenced against either a black perimeter (for absolute calibration), or against the average illumination over the entire field (which would permit color- constancy as in human vision).
Gowling Lafleur Henderson Llp
Zador Andrew M.
LandOfFree
Method, apparatus, and system for extracting denoised,... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method, apparatus, and system for extracting denoised,..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method, apparatus, and system for extracting denoised,... will most certainly appreciate the feedback.
Profile ID: LFCA-PAI-O-1393426