Face Tracking via Occlusion Probability Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face tracking becomes challenging when the face is partially occluded by objects such as sunglasses, masks, or shadows, leading to reduced accuracy in identifying feature points like eyes and mouth, especially in varying luminous environments.
Innovation Solution
A face tracking method that detects the face region, segments it into sub-regions, calculates occlusion probabilities, and adjusts the face model using an occlusion weight map to accurately track the face even when partially occluded, employing feature points, key frames, and probability models to estimate pose and occlusion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face tracking is performed using feature points (eyes, nose, mouth), then face identification accuracy is improved, but tracking reliability deteriorates when the face is occluded by objects or shadows
Solution Approach 1:
The face region is segmented into multiple sub-regions (patches), and occlusion probability is calculated independently for each sub-region. This allows the system to identify which specific parts of the face are occluded and compensate accordingly, maintaining overall tracking reliability even when some features are hidden.
Solution Approach 2:
Different sub-regions of the face are assigned different occlusion probabilities based on local characteristics. The system applies local compensation strategies where occlusion is detected, while maintaining standard tracking where visibility is good, thereby improving overall reliability without sacrificing precision in visible regions.
2Reliability
If the face region is segmented into multiple sub-regions and occlusion probabilities are calculated, then tracking robustness under occlusion is improved, but computational complexity increases
Solution Approach 1:
The system calculates occlusion probability for multiple sub-regions, which is more computation than a single global assessment, but not as intensive as full image analysis. This partial action approach provides sufficient robustness improvement without excessive computational burden.
Solution Approach 2:
The system changes the parameter from a single global face tracking metric to multiple local occlusion probability values for different sub-regions. This parameter transformation enables robustness improvement through localized analysis while managing computational complexity through efficient probability calculations.
3Measurement precision
If occlusion weight maps are used to adjust the face model, then tracking accuracy under occlusion is improved, but processing time increases
Solution Approach 1:
The occlusion weight map is generated in advance based on occlusion probability calculations from sub-regions. This preliminary assessment of occlusion conditions allows the system to quickly adjust the face model without performing complex real-time analysis during tracking, thereby improving accuracy while minimizing additional processing time.
Data Source
AI summary
A face tracking apparatus includes: a face region detector; a segmentation unit; an occlusion probability calculator; and a tracking unit. The face region detector is configured to detect a face region based on an input image. The segmentation unit is configured to segment the face region into a plurality of sub-regions. The occlusion probability calculator configured to calculate occlusion probabilities for the plurality of sub-regions. The tracking unit is configured to track a face included in the input image based on the occlusion probabilities.