Line-of-sight correction using individualized facial data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing line-of-sight detection techniques are influenced by individual differences in facial anatomy and dominant eye usage, leading to inaccurate detection and potential errors in image capturing or alarm triggering.
Innovation Solution
An information processing apparatus that includes a face detection unit, authentication unit, line-of-sight detection unit, and line-of-sight correction unit, utilizing dictionary data to correct for individual differences and accurately determine the line-of-sight orientation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If averaged positional relationships of facial organs or averaged dictionary data are used to detect line of sight, then the detection process is simplified, but detection precision deteriorates due to individual differences in facial anatomy
Solution Approach 1:
The system performs preliminary registration of each user's facial organ positions and line-of-sight characteristics before actual line-of-sight detection. During registration, the user looks at multiple points while the system stores the positional relationships of facial organs (eyes, pupils, nose, mouth) and the corresponding line-of-sight directions. This pre-collected individualized data is then used to correct detection results, eliminating the need to use averaged data that doesn't account for individual differences.
Solution Approach 2:
The system changes the parameters used for line-of-sight detection from fixed averaged values to dynamically adjustable individual-specific parameters. By storing and utilizing user-specific facial organ positions and line-of-sight characteristics registered during the registration phase, the system adapts the detection parameters to match each user's unique facial anatomy, thereby improving detection precision without excessive complexity.
2Device complexity
If dominant eye differences are not considered in line-of-sight detection, then the detection algorithm is simpler, but detection accuracy deteriorates due to shifts in eye orientation
Solution Approach 1:
The system applies different detection parameters and correction values to each eye based on its individual characteristics and the user's dominant eye. During registration, the system separately captures the positional relationship and line-of-sight characteristics of the left and right eyes. The dominant eye's data is given greater weight in the correction process, while the non-dominant eye's data is also utilized with appropriate adjustment. This localized, eye-specific approach accounts for individual differences in eye orientation and dominant eye usage.
3Measurement precision
If individual-specific line-of-sight correction data is registered and used, then line-of-sight detection precision is improved, but system complexity and data storage requirements increase
Solution Approach 1:
The system performs all complex data collection and processing during an initial registration phase, storing individualized facial organ positions and line-of-sight characteristics in advance. During actual line-of-sight detection, the system only needs to retrieve the pre-registered data and apply simple corrections based on the user's current gaze direction. This shifts the computational complexity from the detection phase to the registration phase, making the operational detection process simpler while maintaining high precision.
4Productivity
If averaged dictionary data is used for face and eye orientation detection, then processing speed is maintained, but detection accuracy deteriorates due to individual anatomical variations
Solution Approach 1:
The system changes from using fixed averaged parameters to using dynamically selected individual-specific parameters stored during registration. The registration process captures each user's unique facial organ positions and orientation characteristics, which are then retrieved and applied during detection. This parameter adaptation maintains processing speed by using pre-computed individualized data rather than requiring complex real-time calculations, while simultaneously improving accuracy by accounting for individual anatomical variations.
Data Source
AI summary
An information processing apparatus includes a face detection unit configured to detect a face of a person from an image; a storage unit configured to store dictionary data which holds information relating to faces and line-of-sight correction data corresponding to respective persons; an authentication unit configured to authenticate a person corresponding to the face detected by the face detection unit, using the information relating to faces in the dictionary data; a line-of-sight detection unit configured to detect information relating to a line of sight from the face detected by the face detection unit; and a line-of-sight correction unit configured to correct the information relating to a line of sight detected by the line-of-sight detection unit, using the line-of-sight correction data in the dictionary data corresponding to the person authenticated by the authentication unit.


