Eyewear Face Tracking With Adaptive Sensors for Overlay Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wearable devices, such as smart glasses, struggle with efficient facial detection and tracking, particularly in varying lighting conditions and backgrounds, leading to suboptimal image overlay rendering.
Innovation Solution
A wearable device equipped with a visible light camera and a machine learning model, trained on facial features, tracks positional information to create accurate image overlays using a neural network, optimizing power consumption through ambient light and activity sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If facial detection and tracking is performed using traditional methods in wearable devices, then the device can capture and present images, but the facial detection accuracy deteriorates in varying lighting conditions and backgrounds
Solution Approach 1:
The system changes detection parameters dynamically by switching between different sensor types (visible light camera, infrared camera, depth sensor) based on ambient lighting conditions. In low light, the system transitions from visible light-based detection to infrared or depth-based detection, maintaining accuracy across varying environmental conditions
Solution Approach 2:
The patent introduces intermediate processing layers including machine learning models and multi-sensor fusion algorithms that mediate between raw sensor data and final facial detection results. These intermediaries enhance detection reliability by filtering noise and compensating for limitations of individual sensors in different lighting conditions
2Measurement precision
If continuous facial tracking is performed in real-time, then the system provides accurate image overlays, but power consumption increases
Solution Approach 1:
The system implements periodic facial detection at full resolution followed by continuous tracking at reduced computational cost. After initial face detection, the system uses lighter-weight tracking algorithms that require less processing power, thereby reducing energy consumption while maintaining tracking precision
Solution Approach 2:
The patent dynamically adjusts processing intensity based on detected activity. When facial tracking is stable, the system reduces processing frequency; when tracking quality degrades or lighting changes, processing intensity increases automatically, optimizing the balance between precision and power consumption
3Measurement precision
If multiple sensors and machine learning models are integrated for enhanced facial detection, then detection accuracy improves, but device complexity increases
Solution Approach 1:
The patent segments the facial detection system into distinct functional modules: visible light camera for normal conditions, infrared camera for low light, depth sensor for spatial mapping, and separate machine learning models for each sensor type. This modular segmentation allows independent optimization and simplifies the integration complexity by providing clear interfaces between components
Solution Approach 2:
The system employs a universal processing framework that handles multiple sensor types and detection modes through common algorithms. The machine learning models are designed to process data from different sensor modalities using unified architectures, reducing overall system complexity despite the diversity of input sources
Data Source
AI summary
A wearable or a mobile device includes a camera to capture an image of a scene with a face and a display for displaying an image overlaid on the face. Execution of programming by a processor configures the device to perform functions, including functions to capture, via a camera of an eyewear device, an image of a scene including a face, identify the face in the image of the scene, track positional information of the face with respect to the eyewear device, generate an overlay image responsive the positional information, and present the overlay image on an image display.