Facial Pose Tracking with Adaptive Rigid Prior Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rigid prior models fail to adapt to different facial sizes and expressions, leading to unsatisfactory rigidity stabilization and accuracy in facial pose estimation during real-time face tracking.
Innovation Solution
A method for training an adaptive rigid prior model that initializes model parameters as functions of facial sizes, updates them based on facial data, and assigns weights to facial regions based on real-time facial sizes and displacements to stabilize rigidity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fixed rigid prior model is used for face tracking, then the model structure is simple, but it cannot adapt to different facial sizes and expressions, leading to poor accuracy
Solution Approach 1:
The patent transforms the fixed rigid prior model into a dynamic adaptive model by introducing facial size as a variable parameter. The model parameters are updated based on real-time facial size measurements, allowing the system to adapt to different facial scales and expressions while maintaining reasonable structural complexity.
Solution Approach 2:
The patent changes the parameters of the rigid prior model dynamically based on facial size. By adjusting model parameters according to detected facial size, the system achieves accurate pose estimation across different facial scales without requiring completely different models for each case.
2Adaptability or versatility
If the rigid prior model parameters are updated frequently to adapt to different facial sizes, then the adaptability improves, but the computational time and complexity increase
Solution Approach 1:
The patent performs preliminary face tracking and facial size measurement before updating the rigid prior model parameters. By preparing facial size data in advance and using it to guide parameter updates, the system reduces unnecessary computational iterations and training time while maintaining adaptability to different facial sizes.
3Reliability
If the rigid prior model uses uniform weights for all facial regions, then the model is simple, but it cannot handle facial expressions that cause non-rigid deformations
Solution Approach 1:
The patent applies different weights to different facial regions based on their local characteristics and susceptibility to expression-induced deformations. By assigning region-specific weights rather than uniform weights, the model maintains rigidity stabilization in stable regions while allowing flexibility in expression-prone regions.
Data Source
AI summary
Provided is a method for training an adaptive rigid prior model. The method includes: initializing model parameters of the adaptive rigid prior model; acquiring a plurality of frames of facial data of a same face by face tracking on training video data using the adaptive rigid prior model; updating the model parameters based on the plurality of frames of facial data; determining whether a condition of stopping updating the model parameters is satisfied; stopping updating the model parameters of the adaptive rigid prior model, and acquiring a final adaptive rigid prior model; and returning to the step of acquiring the plurality of frames of facial data of the same face by face tracking on the training video data using the adaptive rigid prior model.