User-Adaptive Activity Recognition for Data Glasses
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing user activity recognition methods for data glasses lack precision and customization, leading to suboptimal performance in everyday usage and traffic safety, as they do not adequately account for user-specific behaviors and contexts.
Innovation Solution
A user-customized method for data glasses that involves a classifying unit trained through a user-specific training step, utilizing both internal and external sensor data, including neural networks or static classifiers, to recognize user activities with high accuracy and adaptability, thereby enhancing recognition precision and convenience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a generic activity recognition method is used, then device complexity is reduced, but measurement precision of user activities deteriorates
Solution Approach 1:
The system performs preliminary user-specific training before actual activity recognition. During this training phase, the classifying unit learns individual user behavior patterns, movement characteristics, and contextual preferences. This preliminary adaptation enables the system to achieve high measurement precision in subsequent recognition tasks without requiring complex real-time adjustments.
Solution Approach 2:
The classifying unit automatically adapts to each user through self-learning mechanisms. The system continuously refines its activity recognition models by analyzing user-specific sensor data patterns, enabling automatic customization without manual intervention. This self-service approach maintains measurement precision while avoiding the complexity of manual configuration systems.
2Measurement precision
If user-specific training is implemented, then activity recognition precision is improved, but use of energy increases
Solution Approach 1:
The energy-intensive training process is performed in advance during a dedicated training phase, rather than continuously during normal operation. Once training is complete, the system uses the learned user-specific models for efficient real-time recognition with minimal energy consumption. This separates the high-energy learning phase from the low-energy inference phase.
Solution Approach 2:
The system employs lightweight machine learning models that can be rapidly trained and deployed. These models are designed to be computationally efficient, allowing for periodic retraining without excessive energy costs. The models are optimized to run on mobile devices with limited power resources, balancing precision requirements with energy constraints.
3Measurement precision
If multiple external sensors are used for training, then activity recognition precision is improved, but device complexity is worsened
Solution Approach 1:
The system utilizes sensors that users already possess in their daily lives, such as smartphones, smartwatches, or fitness trackers. These multi-functional devices serve both as communication tools and as training data sources. By leveraging existing universal devices rather than requiring specialized equipment, the system achieves high measurement precision without increasing the complexity of the data glasses themselves.
Solution Approach 2:
The system uses an intermediary data exchange mechanism where external sensors communicate activity data to the data glasses through standard communication protocols. This intermediary approach allows the system to leverage rich data from multiple sensor sources without directly integrating complex sensor hardware into the glasses, maintaining simplicity while achieving high recognition precision.
4Adaptability or versatility
If user customization is implemented, then adaptability to user behaviors is improved, but ease of operation deteriorates
Solution Approach 1:
The system automatically performs user-specific adaptation through self-learning mechanisms. The classifying unit independently analyzes user behavior patterns, movement characteristics, and contextual preferences without requiring manual configuration. This automated self-service approach achieves high adaptability to individual users while maintaining ease of operation, as users simply need to wear the device and engage in normal activities for the system to learn and adapt.
Data Source
AI summary
A method for a user activity recognition for data glasses. The data glasses include at least one integrated sensor unit. An activity of a user of the data glasses is recognized, via an evaluation of data that are detected by the integrated sensor unit, in at least one user activity recognition step carried out by a classifying unit. It is provided that when the user activity recognition step is carried out, the classifying unit takes into account information that has been ascertained in a (preferably) user-specific training step chronologically preceding the user activity recognition step, by training the classifying unit.