Behavior Recognition Model Refinement via Sensor Data Interpolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Behavior recognition models face limitations due to missing data intervals caused by sensor errors and human errors in labeling, which restrict the reflection of time-series characteristics of behaviors, leading to suboptimal performance.
Innovation Solution
An apparatus that includes a data pre-processing unit for identifying and interpolating missing values, a behavior recognition unit for generating classification results, a data refinement unit for correcting results based on time-series characteristics, and a learning model update unit for synchronizing and optimizing the behavior recognition model, utilizing a database for representative pattern sample data to improve model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If segment data with missing values is used for behavior recognition training, then the number of available training samples is reduced, but using complete data segments would lose time-series behavioral information
Solution Approach 1:
The system performs preliminary data processing by identifying missing values in sensor data before behavior recognition training. It proactively fills missing values using interpolation methods (linear, cubic, or spline interpolation) based on surrounding valid data points, ensuring that complete time-series segments are available for training without losing behavioral information.
Solution Approach 2:
The system introduces an intermediary data processing layer between raw sensor data and behavior recognition training. This intermediary layer uses reference data from wearable devices and interpolation algorithms to generate complete data segments, acting as a mediator that preserves time-series information while making data suitable for training.
2Measurement precision
If behavior recognition is performed on segmented data with missing values, then classification results are limited, but processing continuous data would increase computational complexity
Solution Approach 1:
The system divides continuous sensor data into fixed-time segments for behavior recognition processing. Each segment is independently processed through the behavior recognition model, enabling efficient parallel computation while maintaining temporal context through the segmentation structure.
Solution Approach 2:
The system performs preliminary interpolation to fill missing values in each segment before classification. By pre-processing segments to complete their data structure, the system avoids complex computational methods during actual classification, reducing real-time computational complexity while improving accuracy.
3Adaptability or versatility
If labeled training data from users is used for model training, then the model can learn from real behaviors, but human errors in labeling reduce reliability
Solution Approach 1:
The system implements a feedback mechanism where behavior recognition results are continuously evaluated and used to refine the model. Validation results from multiple segments are aggregated, and the model is retrained with corrected labels, creating a closed-loop system that improves reliability by identifying and correcting human labeling errors through iterative feedback.
Solution Approach 2:
The system performs self-correction by using its own validation results to identify potential labeling errors. When classification results consistently indicate a different behavior pattern across multiple segments, the system automatically flags potential labeling errors and uses these corrected patterns for model retraining, reducing dependence on potentially erroneous manual labels.
Data Source
AI summary
Provided is an apparatus for refining data and improving the performance of a behavior recognition model by reflecting time-series characteristics of a behavior. The apparatus includes: a data pre-processing unit configured to receive training data and real-time data as input, identify a missing value of sensor data, and interpolate the sensor data; a behavior recognition unit configured to, through a behavior recognition model, generate a behavior recognition classification result for the preprocessed real-time data; a data refinement unit configured to correct the behavior recognition classification result to generate a refined dataset; a learning model update unit configured to analyze a similarity of the refined dataset and, based on a result of the analysis, perform learning to generate the behavior recognition model; and an information output unit configured to express a corrected behavior recognition result to a user.