Head-Pose Action Triggering with Monocular Facial Landmark Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for head pose detection require depth information, are complex, rely on neural networks with extensive training data, and are not subject-independent, making them inefficient and resource-intensive.
Innovation Solution
A method using a monocular camera to detect predetermined facial features, compute likelihoods, and trigger actions based on probabilities without depth information, utilizing existing libraries like OpenPose or PoseNet for feature detection, and employing a stability parameter to ensure reliable image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If depth information is used for head pose detection, then measurement precision is improved, but device complexity and resource requirements increase
Solution Approach 1:
The patent extracts and removes the depth information requirement from the head pose detection system. By using only 2D monocular camera images and detecting facial landmarks (eyes, nose, mouth, ears) in two dimensions, the system achieves head pose estimation without needing depth cameras or multi-ocular setups, thus reducing device complexity while maintaining functional capability
Solution Approach 2:
The patent uses 2D image copies of facial features as substitutes for 3D depth information. By detecting and analyzing the positions of facial landmarks in 2D images and computing their spatial relationships, the system creates a virtual 3D representation that enables head pose estimation without actual depth measurement
2Measurement precision
If trained neural networks are used for head pose classification, then measurement precision is improved, but device complexity and training data requirements increase
Solution Approach 1:
The patent implements a self-service approach where the system computes head pose angles directly from detected facial landmark coordinates using geometric calculations. Instead of relying on pre-trained neural networks that require extensive training data and complex infrastructure, the system serves itself by using simple, deterministic mathematical formulas to convert landmark positions into pose angles, eliminating the need for external training resources
Solution Approach 2:
The patent replaces expensive, complex neural network models with simple, computationally inexpensive geometric calculations. The solution uses basic trigonometry and coordinate geometry that can be executed with minimal computational resources, making it suitable for devices with limited processing power and memory
3Measurement precision
If reference pose definition is required for pose tracking, then measurement precision is improved, but adaptability decreases
Solution Approach 1:
The patent implements a dynamic reference system where the reference pose is not fixed but adapts to each individual user. By detecting facial landmarks on the actual user and computing relative positions dynamically, the system adjusts to different face shapes, sizes, and anatomical variations, enabling personalized pose estimation without requiring pre-defined reference poses for each individual
Data Source
AI summary
Disclosed herein is an apparatus comprising a camera and a processing unit operatively coupled to the camera, wherein the processing unit is configured to: receive a sequence of images captured by the camera; process a first image of the received sequence of images to compute respective likelihoods of each of a plurality of predetermined facial features being visible in the first image; compute, from the computed likelihoods, a probability that the first image depicts a predetermined first side of a human head; responsive to at least the computed probability exceeding the predetermined detection probability, trigger performance of a predetermined action.