Road Tunnel Sensor Fusion for Low-False-Alarm Event Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor systems in road tunnels suffer from high false alarm rates and inefficiencies due to decoupled operation, with current data fusion methods like probabilistic and feature-based approaches being unsuitable or labor-intensive, lacking sufficient training data.
Innovation Solution
A method for training event classifiers using machine learning algorithms that allows for efficient sensor fusion in road tunnels by creating an n-dimensional feature vector from sensor data and enabling operator evaluation for classifier training without labeled data, utilizing algorithms like Nearest Neighbor, SVM, Decision Trees, and Random Forests.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple sensor systems are used to monitor road tunnels, then detection coverage is improved, but false alarm rates accumulate due to decoupled operation of individual systems
Solution Approach 1:
The patent merges multiple decoupled sensor systems into a coupled approach where sensor data are integrated and analyzed together. Individual sensor detections are combined to form composite event assessments, allowing the system to distinguish true events from false alarms by cross-validating multiple sensor inputs simultaneously.
Solution Approach 2:
The patent introduces an intermediary classification system that processes sensor data through trained event classifiers. These classifiers act as intermediaries between raw sensor detections and final event determination, using machine learning to evaluate whether detected events represent true conditions or false alarms based on patterns in the sensor data.
2Measurement precision
If probabilistic data fusion methods are used, then detection performance is improved, but precise sensor installation positions and synchronized timestamps are required which are not available in most road tunnel configurations
Solution Approach 1:
The patent changes the parameters used for sensor data fusion from requiring precise physical installation positions and synchronized timestamps to using relative temporal relationships and event-type based classification. The system transforms the problem from one requiring exact spatial-temporal alignment to one that can work with less precise, more readily available sensor data characteristics.
Solution Approach 2:
The patent replaces complex probabilistic models requiring precise sensor metadata with simpler machine learning classifiers that can be trained on available sensor data. This substitution uses more readily implementable classification algorithms that do not depend on difficult-to-obtain precise installation and synchronization information.
3Adaptability or versatility
If feature-based data fusion with machine learning classifiers is used, then event classification capability is improved, but substantial amounts of labeled training data are required which are not easily obtained
Solution Approach 1:
The patent enables the classification system to improve its own performance through continuous operation. As the system processes sensor data and generates classifications, it accumulates training data from these real-world operations, using the generated classifications and sensor patterns to continuously refine and retrain the event classifiers without requiring external labeled datasets.
Solution Approach 2:
The patent implements feedback loops where classification results and sensor data are fed back into the training process. The system uses its own operational data to continuously retrain and improve the event classifiers, creating a self-improving system that reduces its dependency on external labeled training data over time.
4Device complexity
If individual sensor systems operate independently, then system complexity is reduced, but detection reliability decreases due to inability to cross-validate sensor data
Solution Approach 1:
The patent merges individual sensor system operations into a unified coupled approach where multiple sensor inputs are processed together through shared event classifiers. This allows cross-validation of sensor data while maintaining manageable system complexity through standardized classification processes that handle multiple sensor types uniformly.
Data Source
Figure 1~3
AI summary
The invention relates to methods for sensor fusion of sensors (2) in road tunnels comprising the steps of: • Receiving sensor data (S) from a plurality of sensors (2) in a road tunnel over a period of time (t), wherein the sensor data (S) comprise detected events in the road tunnel, wherein n different events are detectable in the road tunnel with the plurality of sensors (2), • Creating an n-dimensional feature vector (V), wherein each dimension of the feature vector (V) corresponds to the number of detections of an event, • Classifying (K) the feature vector (V) by n trained event classifiers (1), • wherein the event classifiers (1) are pre-trained by an operator evaluating the classification (B) as correct or incorrect via an operator interface.