Sample-Mean Data Clustering for Low-Memory Video Frame Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video processing techniques for identifying specific elements, such as human faces, are time-consuming and resource-intensive, particularly in applications like security and streaming video, due to the large number of frames that need to be analyzed and processed.

Innovation Solution

A clustering method that assigns video frames to samples, calculates mean values for each sample, and uses these summaries to reduce memory consumption and processing time by maintaining only sample means, rather than individual observation data, while dynamically determining when to create new clusters or merge existing ones based on distance thresholds.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If all video frames are stored and processed individually for clustering analysis, then clustering accuracy is improved, but memory consumption increases significantly

Engineering Contradiction:
Improveclustering accuracyVSAvoidmemory consumption
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent creates a simplified copy of the cluster data in memory (cluster data structure with aggregated statistics) that represents the essential characteristics of the full video frame data. This copy allows clustering operations to be performed on compressed representations rather than the complete original data, significantly reducing memory requirements while maintaining clustering functionality

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent transforms the data representation from storing complete video frames to storing aggregated statistical parameters (means, standard deviations, frame counts) that characterize the cluster. This parameter transformation enables the system to work with compact numerical summaries instead of large image data, resolving the memory vs. accuracy contradiction

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If thousands of video frames are processed individually, then identification accuracy is improved, but processing time increases significantly

Engineering Contradiction:
Improveidentification accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments the video processing task into two distinct phases: an offline clustering phase that processes all frames to build cluster models, and an online identification phase that uses these models for rapid classification. This segmentation allows computationally intensive operations to be performed once offline, while online processing becomes much faster

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary clustering analysis offline to pre-compute cluster centers, means, and standard deviations from all video frames. This preliminary action creates ready-to-use cluster models that can quickly classify new frames without repeating the full analysis, significantly reducing online processing time while maintaining accuracy

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If complete observation data is maintained for all classified observations, then clustering precision is improved, but device complexity increases

Engineering Contradiction:
Improveclustering precisionVSAvoiddata management complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the essential statistical features (means and standard deviations) from the complete observation data and stores these extracted features in the cluster data structure. By taking out only the necessary information for clustering operations, the system reduces data management complexity while preserving clustering precision

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4540742B1Sampling technique for data clustering
Publication Date: 2026.01.14 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4540742B1 patent drawingFigure 1
  • EP4540742B1 patent drawingFigure 2A
  • EP4540742B1 patent drawingFigure 2B

AI summary

Techniques and solutions are described for clustering, with reduced memory consumption, observations in a data set. A first observation is received. A distance is determined between the first observation and a first cluster. It is determined that the distance satisfies a threshold, and the first observation is added to a first sample of the first cluster. A first set of means is calculated for the first sample using data of the first observations and data of at least a second observation of the first sample. A second set of means is calculated for the plurality of elements for the first cluster using the first set of means for the first sample and at least a third set of means for at least a second sample of the first cluster. An identifier of the first observation is added to a collection of observation identifiers classified in the first cluster.