Activity Recognition via Embedded Feature Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing activity recognition systems face challenges in generalization ability, leading to poor performance in recognizing unseen data and resource inefficiency due to over-optimization and over-generalization issues.
Innovation Solution
An activity recognition method and system that converts sensor data into activity feature values, compares them with embedded feature values, and uses automatic learning and continuous learning with AI to improve generalization ability, reducing resource usage by recording only count and center point values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional artificial intelligence models are used for activity recognition, then optimization ability is improved (good performance in training set), but generalization ability deteriorates (poor performance in test set with unseen data)
Solution Approach 1:
The patent segments the activity recognition process into multiple stages: data collection from sensors, feature extraction to generate activity feature values, comparison with embedded feature values, and recognition decision-making. This segmentation allows the system to process data in manageable steps, improving both optimization and generalization capabilities by focusing on specific aspects at each stage.
Solution Approach 2:
The patent changes the parameters used for activity recognition by converting raw sensor data into activity feature values through feature extraction. This parameter transformation enables the system to work with more meaningful and generalized features rather than raw data, thereby improving generalization ability while maintaining optimization performance.
2Measurement precision
If comprehensive data processing is performed to improve recognition accuracy, then measurement precision is improved, but resource consumption increases
Solution Approach 1:
The patent extracts only the essential activity feature values from the comprehensive sensor data through feature extraction. By taking out only the relevant features needed for recognition rather than processing all raw data, the system achieves accurate activity recognition while significantly reducing computational load and resource consumption.
Solution Approach 2:
The patent creates embedded feature values that serve as templates or copies of typical activity patterns. During recognition, the system compares new activity feature values against these pre-stored embedded features, which requires minimal computational resources while maintaining high recognition accuracy.
Data Source
AI summary
An activity recognition method includes steps of obtaining a plurality of embedded feature values, converting a data set obtained by at least one sensor into an activity feature value, comparing the activity feature value with the embedded feature values to generate a comparison result, and performing an activity recognition according to the comparison result. Therefore, the present invention achieves the advantages of precisely recognizing activities.