Face Position Tracking Using Dual Flexible Spider Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing driver alertness monitoring systems face challenges in accurately tracking face position, especially in dynamic driving conditions and when facial occlusions occur, leading to unreliable alerts due to hand movements or jerky driving.
Innovation Solution
A system utilizing a monocular camera positioned at an angle, employing a deformable part model and dual flexible spider model to track face position, with corner points refinement and optical flow analysis to differentiate valid from invalid motion vectors, and providing alerts based on predefined thresholds for attention detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If CAMSHIFT algorithms relying on skin color are used for face tracking, then tracking efficiency in daylight is improved, but reliability deteriorates under dynamic driving conditions and facial occlusions
Solution Approach 1:
The face tracking system segments the face into multiple key points (landmarks) rather than treating it as a single object. This segmentation allows the system to track individual facial features independently, improving reliability when partial occlusions occur since not all key points need to be visible for tracking to continue.
Solution Approach 2:
The system changes from color-based tracking parameters to geometric parameter-based tracking using key point coordinates and optical flow vectors. This parameter transformation makes the tracking robust to lighting changes and occlusions, as geometric relationships remain stable even when color information is compromised.
2Duration of action of moving object
If optical flow tracking is used to monitor face position, then continuous tracking is improved, but false alerts increase due to hand movements and jerky driving
Solution Approach 1:
The system implements feedback mechanisms by continuously comparing the current face position with the reference position and using optical flow to detect motion patterns. The feedback loop allows the system to distinguish between intentional face movements (indicating alertness) and spurious movements (caused by hand gestures or vehicle motion) through pattern recognition over time.
Solution Approach 2:
The system dynamically adjusts tracking parameters and motion thresholds based on detected driving conditions. By making the tracking system adaptive to varying driving scenarios, it can differentiate between normal driving vibrations and actual driver inattention, reducing false alerts while maintaining continuous monitoring.
3Measurement precision
If a camera positioned in front of the driver is used for face detection, then frontal face detection is improved, but robustness to occlusions and dynamic movements deteriorates
Solution Approach 1:
The system transitions from two-dimensional color-based face detection to three-dimensional geometric modeling using key point coordinates and spatial relationships. This dimensional enhancement provides depth information and structural context that makes the system robust to occlusions, as the 3D face model can be reconstructed even when parts of the 2D image are obscured.
Data Source
Figure 1
Figure 2
Figure 3(a)~3(b)
AI summary
Current technologies detect the alertness of a user using frontal face. However, there are no techniques to track the position of face from a side angle. A method and system for face position tracking of a user and alerting the user is disclosed. The method includes capturing images using a monocular camera and identifying a user's face from the images and storing the user's face as reference face. The user's face is divided into two regions and one or more corner points are identified. A centroid is created in the two regions and the corner points are joined using virtual lines to create a dual flexible spider model. One or more new corner points are created on fulfillment of pre-defined conditions. Angle and magnitude of flow vectors is determined using one or more new corner points and alerts are given to the user based on the angle and magnitude.