Tunable Signal Sampling for Improved Key-Data Extraction

Number: 20220108113
Published: 2022-04-07
Filed: 2020-10-01
Assignee: DISNEY ENTERPRISES INC.
Inventors: Farre Guiu; Miquel Angel, Pernias; Pablo, Martin; Marc Junyent, Aparicio Isarn; Albert
CPC: G06V20/46; G06V10/94; G06V10/82; G06V10/751
Verdict: Set aside generic data extraction, business
Source: Google Patents · FreePatentsOnline

Abstract

A tunable signal sampling system includes a computing platform having a hardware processor and a memory storing a software code that when executed receives a communications signal, identifies data partitions included in the communications signal, performs, using a first predetermined metric, a first set of comparisons each comparing a different sequential pair of the data partitions with each other, and selects, based on the first set of comparisons, a subset of the data partitions as candidate sample partitions of the communications signal. The software code also determines multiple default sample partitions of the communications signal, performs, using a second predetermined metric, a second set of comparisons each comparing a different one of the default sample partitions with a respective one of the candidate sample partitions, and extracts, using a predetermined weighting factor applied to the results of the second set of comparisons, a sample of the communications signal.

Background

BACKGROUND

Signal sampling has a wide variety of applications in a disparate set of technical fields. By way of example, signal sampling is an important technique in noise analysis, biomedical research, video frame analysis, as well as speech coding and other audio signal processing applications, to name a few. Signal sampling is typically performed at a constant rate across the entirety of the signal, with the granularity of the sampling being controlled by the sampling rate. Although such constant rate sampling may be advantageous for some applications, including for signal quality control purposes, it may impose disadvantages on others.

One example application in which constant rate signal sampling may be suboptimal is video frame extraction for use in training an artificial neural network (ANN), as well as for inference by such an ANN. In the process of training an ANN to perform image analysis, for instance, the ANN may be trained using one set of training data to recognize high intensity actions such as fires, explosions, and gunshots, while an entirely different set of training data may be used for training location recognition. The video frames suitable for use in each type of training will typically have significantly different video characteristics, such as pixel intensity, sharpness, and the like. However, because constant rate signal sampling will extract such frames periodically, many potentially useful training samples may be missed entirely. Conse

Claims

1. A tunable signal sampling system comprising: a computing platform including a hardware processor and a memory storing a software code; the hardware processor configured to execute the software code to: receive a communications signal; identify a plurality of data partitions included in the received communications signal; perform, using a first predetermined metric, a first plurality of comparisons, wherein each of the first plurality of comparisons compares a different sequential pair of the plurality of data partitions with each other; select, based on the first plurality of comparisons, a subset of the plurality of data partitions as candidate sample partitions of the communications signal; determine a plurality of default sample partitions of the communications signal; perform, using a second predetermined metric, a second plurality of comparisons, wherein each of the second plurality of comparisons compares a different one of the plurality of default sample partitions with a respective one of the candidate sample partitions; and extract, using a predetermined weighting factor applied to a result of each of the second plurality of comparisons, a sample of the received communications signal. || 11. A method for use by a tunable signal sampling system including a computing platform having a hardware processor and a memory storing a software code, the method comprising: receiving, by the software code executed by the hardware processor, a communications signal; identifying, by the software code executed by the hardware processor, a plurality of data partitions included in the received communications signal; performing, by the software code executed by the hardware processor and using a first predetermined metric, a first plurality of comparisons, wherein each of the first plurality of comparisons compares a different sequential pair of the plurality of data partitions with each other, selecting, by the software code executed by the hardware processor and based on the first plurality of comparisons, a subset of the plurality of data partitions as candidate sample partitions of the communications signal; determining, by the software code executed by the hardware processor and without using the first predetermined metric, a plurality of default sample partitions of the communications signal; performing, by the software code executed by the hardware processor and using a second predetermined metric, a plurality of second comparisons, wherein each of the second plurality of comparisons compares a different one of the plurality of default sample partitions with a respective one of the candidate sample partitions; and extracting, by the software code executed by the hardware processor and using a predetermined weighting factor applied to a result of each of the second plurality of comparisons, a sample of the received communications signal.