Activity Detection Using Symbolic Time Series Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Unattended ground sensors (UGS) face high false alarm rates and limited reliability in detecting and classifying human activities due to inadequate algorithms and environmental conditions, particularly in seismic data with low signal-to-noise ratio and variability, making it challenging to distinguish between activities like walking and digging.
Innovation Solution
A system utilizing a Multi-scale Symbolic Time Series Algorithm (MSTSA) with a Short-Length Symbolic Time-Series Online Classifier (SSTOC) and Probabilistic Finite State Automaton (PFSA) for real-time detection and classification of human and vehicular activities, denoising seismic data, generating symbols, and analyzing transitions between states to accurately classify activities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional detection algorithms are used in unattended ground sensors, then the system is simple and easy to deploy, but the false alarm rate is high and reliability is low
Solution Approach 1:
The patent segments the detection process into multiple stages: signal acquisition, feature extraction, pattern recognition, and decision-making. By dividing the complex detection task into manageable segments, the system achieves higher reliability through multi-stage verification while keeping each individual stage computationally tractable for deployment on resource-constrained sensor nodes.
Solution Approach 2:
The patent transitions from traditional single-dimensional signal analysis to multi-dimensional feature space analysis. By extracting features across multiple dimensions (time, frequency, spatial, and contextual dimensions), the system dramatically improves detection reliability and reduces false alarms, while the dimensional transformation enables more robust pattern recognition despite increased computational requirements.
2Measurement precision
If complex processing algorithms are implemented to improve detection accuracy, then classification accuracy improves, but power consumption increases
Solution Approach 1:
The patent performs preliminary feature extraction and filtering at the sensor node before transmitting data to central processing units. By pre-processing signals and extracting salient features locally, the system reduces the computational burden on power-constrained devices while maintaining high classification accuracy through selective feature transmission and efficient local processing of only the most informative data.
Solution Approach 2:
The patent introduces intermediate processing layers that act as mediators between raw sensor data and final classification decisions. These intermediate layers perform dimensionality reduction, feature selection, and preliminary classification, thereby reducing the energy required for full-scale processing while preserving the information necessary for accurate activity classification through hierarchical processing stages.
3Reliability
If seismic sensors are used for personnel detection, then detection reliability improves compared to acoustic sensors, but the system becomes more sensitive to environmental variations like soil type and moisture
Solution Approach 1:
The patent dynamically adjusts detection parameters such as threshold values, filtering characteristics, and feature extraction parameters based on environmental conditions. By monitoring environmental variables like soil type and moisture content and adapting the detection parameters accordingly, the system maintains high detection reliability across varying environmental conditions through real-time parameter optimization and environmental compensation techniques.
Data Source
AI summary
A system for detection of human or vehicle activity comprising at least one sensor adapted to generate a signal and at least one processor operating to denoise the signal; generate an autocorrelation of the signal; partition the signal into a predetermined number of overlapping segments to form a time series of data; generate symbols for the overlapping segments; compare the pattern of generated symbols with known predetermined patterns of symbols representing human or vehicular activity; determine whether a threshold probability is exceeded which attributes the data signal to human or vehicular activity; analyze the patterns presented in the data signal by transforming the patterns of symbols into states; determine the transitions between states; and classify the signal as to being attributable to human or vehicular activity based upon the transitions between states. A method of detection and classification of sensor data signals via detecting patterns using time series analysis.