Landmark Database for Face Direction-Corrected Action Unit Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems fail to accurately detect action units on human faces due to variations in face direction, leading to incorrect determination of action units.
Innovation Solution
The system corrects landmark distances based on face direction angles to improve detection accuracy by generating a landmark database that associates action units with facial parts and using selected landmarks to create face data representing virtual human characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If landmark detection is performed without considering face direction, then the detection process is simple, but the detection accuracy deteriorates due to variations in face direction
Solution Approach 1:
The patent applies parameter changes by adjusting landmark detection based on face direction angles. The system detects face direction from the face image and uses this information to correct landmark positions or select appropriate landmarks, thereby maintaining high detection accuracy across different face orientations without overly complicating the detection process
Solution Approach 2:
The patent implements preliminary action by pre-establishing a landmark database that categorizes landmarks by facial parts and associates them with action units. This pre-organized database structure allows the system to quickly retrieve and select appropriate landmarks based on detected face direction, improving accuracy without significant computational overhead during actual detection
2Adaptability or versatility
If a comprehensive landmark database is generated, then the learning model can access diverse data, but the data accumulation process becomes more complex
Solution Approach 1:
The patent applies segmentation by dividing the landmark database into organized categories based on facial parts (e.g., eyes, nose, mouth, cheeks) and further subdividing by action units. This structured segmentation makes the data accumulation process more manageable and allows systematic organization of diverse face data without creating an unmanageably complex accumulation process
Solution Approach 2:
The patent implements universality by creating a landmark database that serves multiple functions: it stores landmark positions, associates landmarks with action units, categorizes by facial parts, and supports various detection tasks. This multi-functional database structure generates diverse face data while consolidating multiple processing functions into a single systematic approach
Data Source
AI summary
A data accumulation apparatus has: a detecting device that detects, based on a face image in which a face of a human is included, a first landmark of the face; an obtaining device that obtains a state information indicating a state including at least one of an action unit, an emotion and a physical condition of the human included in the face image; and a database generating device that generates a landmark database that includes first landmarks with each of which the state information is associated and which are categorized by a unit of the facial part. A data generation apparatus has: a selecting device that selects at least one first landmark with which the state information indicating that the state is a desired state is associated for each facial part as a second landmark from the landmark database; and a face data generating device that generates a face data that represents a characteristic of a face of a virtual human by using second landmarks by combining the second landmarks that correspond to the facial parts, respectively.