Activity Classification Embeddings for Unknown Motion Rejection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human activity classification systems struggle with accurately identifying and rejecting unknown activities, particularly in open-set scenarios, and are susceptible to sensor artifacts and environmental uncertainties.
Innovation Solution
A Bayesian inference framework using a Kalman filter and a quadruplet loss-based embedding model to generate and track embedding vectors, combined with classification gating and a linear classifier for precise activity classification, effectively handling unknown activities and reducing false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional activity classification methods are used, then the system is simpler to implement, but classification accuracy deteriorates in environments with sensor artifacts and uncertainties
Solution Approach 1:
The patent introduces an intermediary embedding layer that transforms raw motion sensor data into embedded representations. This embedding layer acts as a mediator between the sensor data and classification algorithms, extracting meaningful features while filtering out sensor artifacts and uncertainties, thereby improving classification accuracy without directly increasing the complexity of the entire system
Solution Approach 2:
The patent replaces traditional mechanical signal processing approaches with data-driven embedding models that learn optimal representations from data. Instead of using fixed filtering mechanisms, the system uses neural network-based embedding layers that automatically adapt to different activity patterns and sensor noise characteristics, achieving higher accuracy while maintaining computational efficiency
2Reliability
If the system is more robust to sensor artifacts, then reliability improves, but the ability to detect and measure unknown activities deteriorates
Solution Approach 1:
The patent implements feedback mechanisms where the embedding layer continuously learns from the classification results and adjusts its representations. The system uses feedback from both known and unknown activity classifications to refine the embedding space, allowing it to maintain robustness to sensor artifacts while improving its ability to detect and measure unknown activities over time
Solution Approach 2:
The patent dynamically changes the parameters of the embedding space based on the characteristics of detected activities. When unknown activities are detected, the system adjusts the embedding space parameters to accommodate these new patterns, thereby maintaining both robustness to sensor artifacts and the ability to detect unknown activities
Data Source
AI summary
One or more computing devices, systems, and/or methods are provided. In an example, a method comprises receiving, by a device, incoming motion data from a motion sensor, generating, by the device, an incoming embedding vector based on the incoming motion data, generating, by the device, a predicted embedding vector based on the incoming embedding vector, assigning, by the device, an activity classification based on the predicted embedding vector, and modifying an operating parameter of the device based on the activity classification.