Egocentric Gesture Recognition with Multi-Sensor Posture Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gesture recognition systems face limitations in field of view and viewpoint, leading to increased depth ambiguity and severe occlusion, particularly with egocentric vision sensors, which affect accurate and stable posture estimation.
Innovation Solution
A posture estimation apparatus utilizing two sensors mounted on a user's body, one facing downwards and the other aligned with the user's line of sight, to capture complementary images, and an improvement unit that generates refined posture information based on the change amounts in these images, combined with external sensor data to enhance estimation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If egocentric vision sensors are used to overcome field of view limitations, then the sensor can capture user movements more effectively, but viewpoint limitations cause increased depth ambiguity and severe occlusion
Solution Approach 1:
The patent combines multiple sensors (first sensor and second sensor) with different shooting directions to capture complementary images. The first sensor captures images in a first direction while the second sensor captures images in a second direction, merging their data to overcome the limitations of single-sensor egocentric vision systems and reduce depth ambiguity and occlusion issues
Solution Approach 2:
The patent introduces a temporal dimension by capturing images at different time points and comparing change amounts between frames. This temporal analysis adds another dimension to the spatial data, enabling more accurate posture estimation by detecting movements and changes over time that single-frame analysis would miss
2Measurement precision
If multiple sensors with different shooting directions are used, then posture estimation accuracy is improved, but device complexity increases
Solution Approach 1:
The patent makes each sensor serve multiple functions: the first sensor and second sensor not only capture images for posture estimation but also provide change amount information for improvement unit processing. The system architecture allows the same hardware components to fulfill multiple roles in the posture estimation pipeline, reducing the need for additional specialized devices
Solution Approach 2:
The improvement unit processes the raw sensor data and change amount information to generate improved posture estimation information autonomously. The system self-corrects and refines its own estimations by comparing change amounts across different sensors and time points, reducing the need for external calibration or additional complex processing hardware
Data Source
AI summary
Various embodiments of human motion recognition technology in an egocentric capture environment are disclosed. In one embodiment, a posture estimation apparatus comprises: a first estimation unit configured to estimate a user's posture based on a first image obtained through a first sensor attached to the user, the first image comprising an image of at least a portion of the user, and to generate first information representing the estimated posture; and an improvement unit configured to generate second information by improving the first information based on a second image obtained through a second sensor having a different shooting direction from the first sensor.