Human Action Recognition Using Body and Target Region Fusion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Computer vision systems struggle to identify human actions with respect to small objects in near real-time due to the need for high resolution images, which increases processing time and is infeasible for applications like augmented reality, robotics, and security.

Innovation Solution

A system combines lower resolution images of the body with higher resolution images of targeted body parts using a single vision model to process both, allowing for efficient identification of human actions by resizing and dividing the images into non-overlapping patches with different resolutions, and using positional embeddings to capture detailed information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If high resolution images are used to identify small objects, then identification accuracy is improved, but processing time increases

Engineering Contradiction:
Improveobject identification accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent divides the image processing task into two segments: a first vision model processes the entire image at lower resolution to identify potential regions, while a second vision model processes only the identified region at higher resolution for detailed object identification. This segmentation allows the system to maintain high identification accuracy for small objects while reducing overall processing time by avoiding full high-resolution image processing.

Inventive Principle:
Principle #1Segmentation

2Device complexity

If a single vision model is used to process both body and target images, then device complexity is reduced, but processing efficiency decreases

Engineering Contradiction:
Improvenumber of vision modelsVSAvoidprocessing efficiency
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The patent segments the vision processing task into two distinct models with specialized functions: a first vision model for scanning the entire image at lower resolution to locate regions of interest, and a second vision model for detailed analysis of the target region at higher resolution. This functional segmentation improves processing efficiency by allowing each model to be optimized for its specific task and by enabling parallel processing of different image regions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a dimensional distinction in processing by applying different resolution levels and processing depths to different parts of the image. The first vision model operates on the full image dimension at lower resolution, while the second vision model focuses on a subset dimension (the identified region) at higher resolution, creating a multi-dimensional processing approach that balances complexity and efficiency.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4700713A1Combining body and target regions for identification of a human action with respect to an object
Publication Date: 2026.02.25 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4700713A1 patent drawingFigure 1
  • EP4700713A1 patent drawingFigure 2
  • EP4700713A1 patent drawingFigure 3A

AI summary

A system uses a single vision model to combine lower resolution images of a body and higher resolution images of a targeted body part to more efficiently identify a human action with respect to an object. The system receives images of a scene that include a body. For instance, the images may be sequential frames in a video captured by a camera. The system generates a body image by extracting a region from an image that includes a body. The system generates a target image by extracting a region from the image that includes a targeted body part interacting with an object. The system is configured to perform similar operations on the body image and the target image to ensure that a single vision model can process the target image at a more granular level compared to the body image.