Face Tracking Using Template Normalization and GPU Views
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer systems face challenges in efficiently tracking a user's face across multiple video frames, especially when the face is partially obscured or undergoes changes in orientation and scale, leading to potential tracking errors and increased computational burden.
Innovation Solution
A method where a computer system generates and compares candidate templates of a user's face across frames, normalizing them for size and orientation, and using a graphics processing unit to create multiple views, allowing for efficient face tracking without performing face detection on every frame.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face detection is performed on every video frame, then tracking accuracy is improved, but computational burden increases
Solution Approach 1:
The system performs face detection only on selected frames (e.g., every Nth frame or key frames) rather than all frames, preparing template images in advance. This preliminary action reduces the frequency of computationally intensive detection operations while maintaining tracking accuracy through template matching on remaining frames.
Solution Approach 2:
The system creates template images from detected face regions and stores them for later comparison. Instead of performing full face detection on every frame, the system copies and reuses these templates to match against subsequent frames, significantly reducing computational burden while maintaining tracking precision.
2Reliability
If the system processes every video frame for face tracking, then tracking reliability is improved, but processing time increases
Solution Approach 1:
The system implements periodic face detection at intervals rather than continuously processing every frame. By detecting faces periodically and using template matching for intermediate frames, the system maintains reliable tracking while reducing processing time and computational resources required.
Solution Approach 2:
The system skips intermediate frames that do not require full face detection processing. By rushing through non-critical frames using faster template matching methods instead of comprehensive face detection, the system reduces overall processing time while maintaining tracking reliability through strategic detection at key moments.
3Measurement precision
If the system uses comprehensive face detection algorithms, then detection accuracy is improved, but device complexity increases
Solution Approach 1:
The system segments the video processing task into two distinct phases: comprehensive face detection on selected frames to create templates, and simpler template matching on subsequent frames. This segmentation allows the system to use accurate detection algorithms only where necessary while employing simpler, faster matching algorithms for routine processing, thereby reducing overall system complexity.
Solution Approach 2:
The system creates simplified template representations of faces and uses these copies for rapid matching comparisons. By replacing complex real-time detection algorithms with template copying and comparison, the system maintains high detection accuracy for critical moments while significantly reducing algorithmic complexity for continuous processing.
Data Source
AI summary
In general, the subject matter described in this specification can be embodied in methods, systems, and program products. A computing system accesses an indication of a first template that includes a region of a first image. The region of the first image includes a graphical representation of a face. The computing system receives a second image. The computing system identifies indications of multiple candidate templates. Each respective candidate template from the multiple candidate templates includes a respective candidate region of the second image. The computing system compares at least the first template to each of the multiple candidate templates, to identify a matching template from among the multiple candidate templates that includes a candidate region that matches the region of the first image that includes the graphical representation of the face.