Multi-Pose Face Tracking via Hybrid Color and Edge Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face tracking in surveillance videos is challenging due to similar background colors, ambient illumination variations, and large pose changes, especially when faces are small and partially occluded, requiring robust methods to maintain accurate tracking.
Innovation Solution
A system that combines a color model and an edge model, using Difference of Gaussian (DOG) filters and YCbCr color space, to predict face positions and track faces across frames, incorporating a particle filter for robustness and adaptability to illumination and pose changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If color model is used for face tracking, then invariance to head pose variations is achieved, but tracking fails when background is of similar color or under ambient illumination variations
Solution Approach 1:
The patent combines color-based appearance model with edge-based appearance model to create a hybrid tracking system. The color model provides pose invariance while the edge model provides robustness to illumination variations and background discrimination, resolving the contradiction between pose adaptability and illumination reliability
Solution Approach 2:
The patent uses a composite appearance model that integrates multiple feature types (color histograms and edge information) similar to composite materials. This composite model leverages the strengths of each individual model while compensating for their weaknesses, achieving both pose invariance and illumination robustness
2Reliability
If edge information is used for face tracking, then robustness to illumination variations is achieved, but out-of-plane pose variations worsen 2D edge model matching
Solution Approach 1:
The patent merges edge-based appearance model with color-based appearance model in a unified tracking framework. The edge model provides illumination robustness while the color model handles pose variations, resolving the contradiction between illumination reliability and pose adaptability
Solution Approach 2:
The patent dynamically adjusts the weighting or contribution of different appearance models based on detected conditions such as pose angle and illumination levels. When pose changes are detected, the system increases reliance on the color model, while under illumination variations it relies more on edge information
3Measurement precision
If local features are detected and tracked, then face orientation can be computed, but tracking becomes difficult when faces are small (24×24 pixels) or partially occluded
Solution Approach 1:
The patent transitions from relying solely on local geometric features to incorporating global color distribution features. Color histograms provide a dimensionality-reduced representation that is robust to occlusion and scale changes, while still enabling orientation estimation through color pattern analysis
4Ease of operation
If global features such as color histogram are used, then tracking is simplified, but tracking fails under illumination changes and with cluttered backgrounds
Solution Approach 1:
The patent combines simple global color histogram features with edge-based features to maintain tracking simplicity while improving reliability. The edge features provide discrimination power against cluttered backgrounds and illumination variations without significantly increasing computational complexity
Data Source
AI summary
A system and method are provided for tracking a face moving through multiple frames of a video sequence. A predicted position of a face in a video frame is obtained. Similarity matching for both a color model and an edge model are performed to derive correlation values for each about the predicted position. The correlation values are then combined to determine a best position and scale match to track a face in the video.