Head Pose Estimation Using Accessory Pose Under Landmark Occlusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing head pose prediction systems are hindered by the obscuration of face landmark features due to head accessories, leading to degraded accuracy and feasibility, and existing solutions involving wearable markers or infrared light emitters are inconvenient and cumbersome.
Innovation Solution
A system and method that utilize both face landmark features and head accessory poses to enhance head pose prediction accuracy, employing tracking cameras and processors to determine the position and type of head accessories, and integrate them with face landmark detection to improve head pose estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If head accessories are worn to protect or style the user, then user comfort and style are improved, but face landmark features are obscured leading to degraded head pose prediction accuracy
Solution Approach 1:
The system introduces head accessory pose detection as an intermediary element. Instead of directly relying on obscured face landmarks, the system detects the pose of the head accessory (hat, glasses, headset) and uses it as a mediator to infer the head pose. The processor determines head accessory pose from image data and combines it with visible face landmark features to calculate the final head pose, effectively using the accessory as a surrogate marker when landmarks are obscured.
Solution Approach 2:
The system replaces the traditional mechanical/optical reliance on visible face landmarks with a computational approach. The processor uses image processing and pose estimation algorithms to detect head accessory positions and orientations, then mathematically combines this with partial face landmark data to derive head pose, substituting the direct visual measurement approach with a computational fusion method.
2Measurement precision
If wearable markers or infrared light emitters are used to track head pose, then head pose tracking accuracy is improved, but device complexity and user convenience are worsened
Solution Approach 1:
The system makes the head accessory itself serve the dual purpose of its original function (protection, style) and as a tracking marker. The existing head accessory worn by the user is detected and used for pose estimation, eliminating the need for separate wearable markers or infrared emitters. The accessory essentially serves itself as the tracking element.
Solution Approach 2:
The head accessory is given multiple functions: its original protective or stylistic function, and additionally serves as a pose tracking marker. The same object (hat, glasses, headset) that the user wears for comfort or style purposes is simultaneously utilized by the system for head pose detection, eliminating the need for dedicated tracking devices.
3Measurement precision
If more face landmark features are detected for accurate head pose prediction, then head pose prediction accuracy is improved, but the system becomes more sensitive to obscuration by head accessories
Solution Approach 1:
The system accepts that only partial face landmark features may be visible due to obscuration. Instead of requiring complete landmark detection, the system works with the available partial landmark data combined with head accessory pose information. The processor fuses partial face landmark features with head accessory pose to achieve reliable head pose estimation even when full landmark detection is not possible.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A set of images of a user are captured during a given time period, by employing at least one tracking camera. In each image, at least one face landmark feature of a face of the user is detected, and a position of the at least one face landmark feature is determined. When it is detected in a given image that the user is wearing at least one head accessory (204), a pose of the at least one head accessory in the given image is determined. A pose of a head (202) of the user in the given image is then determined, based on the position of the at least one face landmark feature detected in the given image and the pose of the at least one head accessory in the given image.