Video Face Identity Tracking Through Filtered Tracklet Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for tracking identities in videos face challenges such as individuals appearing and disappearing, drastic appearance changes, blur, occlusions, and lighting variations, leading to inconsistent results and computational inefficiencies.
Innovation Solution
A system that employs object tracking to generate tracklets from consecutive frames, filters noisy face feature vectors, and clusters these tracklets to associate different occurrences of individuals with unique identities, leveraging temporal information and allowing for parallel processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If face feature vectors from all detected faces are used for clustering, then more tracking data is available, but computational resources and processing time increase significantly
Solution Approach 1:
The patent extracts and removes noisy face feature vectors from the clustering process. By identifying and excluding low-quality detections (blurry, occluded, poorly-lit faces), the system reduces the computational burden while maintaining tracking accuracy through a filtered subset of reliable features
Solution Approach 2:
The patent segments the face detection and feature extraction process into quality assessment stages. By dividing the processing pipeline to first detect faces, then evaluate quality metrics, and finally select only high-quality features for clustering, the system optimizes computational efficiency
2Adaptability or versatility
If all detected faces are included in clustering, then comprehensive tracking is achieved, but noisy data from poor quality faces degrades tracking precision
Solution Approach 1:
The patent extracts and removes noisy face feature vectors from the clustering process. By identifying and excluding low-quality detections (blurry, occluded, poorly-lit faces), the system reduces the computational burden while maintaining tracking accuracy through a filtered subset of reliable features
Solution Approach 2:
The patent applies local quality assessment to individual face detections by evaluating specific quality metrics (sharpness, illumination, occlusion level) for each detected face. This allows differential treatment of features based on their local quality characteristics
3Duration of action of moving object
If face feature extraction is performed on all frames, then complete temporal information is captured, but processing latency increases
Solution Approach 1:
The patent performs preliminary quality assessment of face detections before extracting and using feature vectors for clustering. By pre-evaluating quality metrics and filtering out low-quality detections in advance, the system reduces the effective processing workload while maintaining comprehensive temporal coverage through the filtered feature set
Data Source
AI summary
Some aspects of the technology described herein perform identity identification on faces in a video. Object tracking is performed on detected faces in frames of a video to generate tracklets. Each tracklet comprises a sequence of consecutive frames in which each frame includes a detected face for a person. The tracklets are clustered using face feature vectors for detected faces of each tracklet to generate a plurality of clusters. Information is stored in an identity datastore, including a first identifier for a first identity in association with an indication of frames from tracklets in a first cluster from the plurality of clusters.