Autoencoder Motion Detection for Abnormal Activity Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Detecting abnormal motions in repetitive human activities from sensor data is challenging due to large motion variations and data drift, which can lead to false positive detections and invalid comparisons with standard motions, especially in scenarios like fitness monitoring and manufacturing.
Innovation Solution
An autoencoder neural network framework is used to fuse multiple sensors and modalities, generating a latent motion space for standardization evaluation and outlier detection, where each motion is compared to a set of predefined standards and other recorded motions to identify true abnormalities, providing robustness against motion variation and data drift.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If motion data is compared directly with standard motions, then detection simplicity is maintained, but false positive detections increase due to motion variations and data drift
Solution Approach 1:
The patent introduces an autoencoder neural network as an intermediary between raw motion data and abnormality detection. The autoencoder learns a latent motion space that transforms variable motion data into a standardized representation, enabling reliable comparison while maintaining system simplicity. This mediator handles the complexity of motion variations and data drift internally, so users don't need to manually manage these complexities.
Solution Approach 2:
The patent transforms motion data from raw sensor space to latent motion space through the autoencoder's encoding process. This parameter transformation changes the representation of motion data into a form that is invariant to common variations and drift, allowing for more reliable abnormality detection without increasing user-facing complexity.
2Reliability
If autoencoder neural network with latent motion space is used, then detection accuracy improves by reducing false positives, but system complexity increases
Solution Approach 1:
The autoencoder is pre-trained offline to learn the latent motion space and normal motion patterns before deployment. This preliminary action captures the complexity of motion variations and data drift in the trained model weights, so the deployed system can perform accurate detection without needing to process this complexity in real-time, reducing operational complexity.
Solution Approach 2:
The patent uses the autoencoder to create a copied representation of motion data in latent space that preserves essential characteristics while eliminating noise and variations. This copied representation in the latent motion space enables accurate comparison with standard motions without dealing with the complexities of raw sensor data variations.
3Ease of operation
If only standardization evaluation is used, then comparison with standard motions is straightforward, but incomplete reference data leads to invalid comparisons
Solution Approach 1:
The patent implements a two-stage detection process where outlier detection feedback refines the abnormality identification. The system first performs standardization evaluation, then uses outlier detection to validate findings against the learned latent motion space distribution. This feedback mechanism ensures that comparisons are valid even when reference data is incomplete, as the autoencoder learns from available data to establish normal variation patterns.
Data Source
AI summary
Abnormal motions are detected in sensor data collected with respect to performance of repetitive human activities. An autoencoder network model is trained based on a set of standard activity. Repetitive activity is extracted from sensor data. A first score is generated indicative of distance of a repetition of the repetitive activity from the standard activity. The repetitive activity is used to retrain the autoencoder network model, using weights of the autoencoder network model as initial values, the weights being based on the training of the autoencoder network model using the set of standard activity. A second score is generated indicative of whether the repetition is an outlier as compared to other repetitions of the repetitive activity. A final score is generated based on a weighting of the first score and the second score.