Video Person Identification Using Emotion-Aware Graph Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current facial recognition technologies struggle to accurately identify individuals in diverse video contexts due to varying facial physiognomy, age, and emotional states, exacerbated by global interactions and advanced data processing technologies that can generate counterfeit videos.
Innovation Solution
A system and method utilizing STEM-IDR (Spatio-Temporal-Emotion-Mesh Data Identification Recognition) that extracts and processes facial and emotional data through a fully adaptive graph convolutional network (FAGC) to generate unique person representation vectors, incorporating anatomical landmarks and emotional states, and transition probabilities to enhance identification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional facial recognition methods are used, then the system is simple to implement, but identification accuracy deteriorates in diverse video contexts including different emotions, ages, and physiognomies
Solution Approach 1:
The patent segments facial identification into multiple independent components: anatomical landmark detection, emotional state recognition, and identity verification. Each component processes specific features separately before integration, allowing the system to handle diverse facial expressions and physiognomies without compromising accuracy while maintaining manageable system complexity through modular architecture.
2Reliability
If facial recognition focuses only on anatomical landmarks, then the system is computationally efficient, but identification reliability deteriorates when emotional states or age variations are present
Solution Approach 1:
The patent merges anatomical landmark features with emotional state features into a unified identification model. By combining these complementary feature sets, the system achieves higher reliability across diverse conditions (different emotions, ages, physiognomies) while optimizing computational energy through shared processing components and feature fusion strategies.
3Measurement precision
If the system processes only static facial images, then processing speed is high, but the ability to detect emotional nuances and temporal patterns deteriorates
Solution Approach 1:
The patent extracts and processes only the most discriminative temporal features from video sequences rather than analyzing all frames in detail. By identifying and focusing on key frames that capture emotional transitions and temporal patterns, the system achieves high emotional state detection precision while minimizing overall processing time through selective frame analysis.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Systems, methods, and computer readable media for identifying a person in a video are disclosed. Systems, methods, devices, and non-transitory computer readable media may include at least one processor that may be configured to generate a spatiotemporal emotion data compendium (STEM-DC) from the video and to process the STEM-DC using a deep fully adaptive graph convolutional network (FAGC) to determine a first person representation vector that represents the person in the video.