Video Face Recognition via Temporal Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition methods in videos suffer from low accuracy due to the reliance on independent frame comparisons, which are affected by low-quality frames and mismatched face angles, leading to inaccurate detection results.
Innovation Solution
A method and device for face recognition in videos that perform feature extraction on multiple frames to generate face feature vectors, followed by time sequence feature extraction to convert these vectors into a predetermined dimension, utilizing a classifier for recognition, leveraging the temporal correlation between frames to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If independent frame comparison is used for face recognition, then the recognition process is simple and fast, but the recognition accuracy is low due to low-quality frames and mismatched face angles
Solution Approach 1:
The patent merges multiple frame features into a unified feature vector by performing feature extraction on multiple frames containing the same target face and combining their feature vectors. This integration of temporal information from multiple frames improves recognition accuracy while maintaining a manageable processing framework through systematic feature fusion.
Solution Approach 2:
The patent introduces a temporal dimension to the recognition process by extracting features across multiple frames rather than analyzing single frames in isolation. This dimensional expansion from spatial-only to spatio-temporal feature extraction enables the system to leverage temporal correlations and improve accuracy despite increased processing complexity.
2Measurement precision
If multiple image frames are processed to improve recognition accuracy, then the recognition accuracy improves, but the processing time and computational complexity increase
Solution Approach 1:
The patent performs preliminary feature extraction on multiple frames to generate feature vectors before the final recognition decision. By pre-processing and extracting features from all relevant frames in advance, the system prepares comprehensive temporal information that can be efficiently utilized during recognition, reducing the need for repeated processing.
Solution Approach 2:
The patent maintains continuous feature extraction across multiple frames to capture temporal variations and correlations. This continuous processing approach ensures that useful information from each frame is captured and integrated, improving recognition accuracy while the systematic methodology keeps processing efficiency manageable through consistent feature fusion.
Data Source
AI summary
A method for face recognition in the video comprises: performing feature extraction on a target face in multiple image frames in the video to generate multiple face feature vectors respectively corresponding to the target face in the multiple image frames; performing time sequence feature extraction on the plurality of face feature vectors to convert the plurality of face feature vectors into a feature vector of a predetermined dimension; and judging the feature vector of the predetermined dimension by using a classifier so as to recognize the target face.