Gesture Recognition With Camera Motion Compensation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Gesture recognition systems fail in robotic applications with mobile cameras due to coupled hand and camera motions, leading to blurred hand images and classification errors, especially when the hand is far from the camera.

Innovation Solution

A generative model using a motion disentangle generator and hand image generator, combined with multi-stream gesture recognition and gate attention, compensates for camera motion and enhances hand image quality for robust gesture recognition.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a camera is attached to a mobile robot for gesture recognition, then the robot can operate in large and free space, but hand and camera motions become coupled leading to classification errors

Engineering Contradiction:
Improverobot mobilityVSAvoidgesture classification accuracy
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent segments the motion analysis into two independent components: camera motion (estimated from sequential frames) and hand motion (extracted after compensation). This separation allows the system to handle mobile robot applications while maintaining accurate gesture recognition by processing each motion component independently.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary compensation mechanism that estimates camera motion from sequential image frames and uses this estimation to correct subsequent frames. This intermediary step decouples the hand motion from camera motion, enabling reliable gesture recognition in mobile robot scenarios.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If the hand is far from the camera in mobile robot applications, then the robot can operate in large space, but hand images become blurred and difficult to detect

Engineering Contradiction:
Improveoperating space rangeVSAvoidhand detection accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent performs preliminary motion compensation by estimating camera motion from the current and previous frames before hand detection and recognition. This preliminary action of correcting for camera motion ensures that hand images remain clear and detectable even when the hand is far from the camera, maintaining measurement precision across large operating spaces.

Inventive Principle:
Principle #10Preliminary action

3Speed

If camera motion is strong in mobile robot applications, then the robot can move freely, but hand images are easily blurred due to camera motion and distance

Engineering Contradiction:
Improvecamera mobilityVSAvoidimage blurring
Core Design Contradiction:
SpeedVSObject-affected harmful factors

Solution Approach 1:

The patent implements a feedback mechanism where the current frame is used to estimate camera motion, which then feeds back to compensate the next frame. This continuous feedback loop allows the system to maintain clear hand images despite strong camera motion, as each frame's motion information is used to correct the subsequent frame.

Inventive Principle:
Principle #23Feedback

Data Source

PatentUS12633114B2Methods and devices for gesture recognition
Publication Date: 2026.05.19 INTEL CORP
  • US12633114B2 patent drawing
  • US12633114B2 patent drawing
  • US12633114B2 patent drawing

AI summary

Disclosed herein is a gesture recognition device that includes an input interface configured to receive a sequence of images, each image showing a body part with which a gesture is performed from a viewpoint of a camera. The gesture recognition device also generates a sequence of motion-compensated images from the sequence comprising generating a motion-compensated image for an image of the sequence by compensating the movement of the camera viewpoint from a reference camera viewpoint to the viewpoint from which the image shows the body part based on the image and a motion-compensated image of the sequence generated for a preceding image of the sequence which precedes the image in the sequence and estimate the gesture from the sequence of motion-compensated images.