Vehicle Speed Estimation Using IMU Windows and Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile devices rely heavily on GPS receivers for vehicle speed estimation, which consume significant power and reduce battery life, making them undesirable for continuous use in vehicle crash detection.
Innovation Solution
Utilize an Inertial Measurement Unit (IMU) sensors, such as accelerometers and gyroscopes, to estimate vehicle speed by converting sensor measurements into a frequency domain, filtering high frequencies, and defining contiguous windows to generate speed estimates, reducing power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If vehicle speed is measured using traditional sensor fusion methods, then the measurement process becomes complex requiring multiple sensors and computational algorithms, but the measurement precision and reliability remain insufficient
Solution Approach 1:
The patent replaces traditional mechanical sensor fusion systems with a deep learning-based neural network system. Instead of using multiple physical sensors (accelerometers, gyroscopes, barometers) combined with complex fusion algorithms, the system uses a neural network that processes sensor data through learned patterns, achieving more accurate speed measurements with simplified system architecture
Solution Approach 2:
The patent transforms the approach by changing from direct sensor value fusion to using a neural network that learns optimal parameter combinations from training data. The system converts raw sensor measurements into speed predictions through trained weight matrices and bias terms, allowing the system to adapt to different driving conditions and sensor characteristics
2Measurement precision
If deep learning models are trained on large datasets to improve speed prediction accuracy, then the model performance improves, but the training time and computational resources increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network model offline using large datasets before deployment. The model learns optimal weight matrices and bias terms in advance during a training phase, so that during actual vehicle operation, only inference is performed rather than full training. This separates the time-consuming training process from the real-time operation
Solution Approach 2:
The patent segments the development process into distinct phases: offline training phase where the neural network learns from large datasets to create optimized weight matrices, and online inference phase where the pre-trained model processes sensor data in real-time. This segmentation allows comprehensive training without impacting real-time performance
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Apparatuses and methods for predicting a crash using estimated vehicle speed. A set of sensor measurements are received from a mobile device disposed within a vehicle. A set of contiguous windows based on the sensor measurements may be defined. Each contiguous window represents a contiguous portion of the sensor measurements. A set of sensor measurements may be defined for each contiguous window. A trained neural network may execute, using the set of features, to generate one or more speed predictions. A vehicle crash prediction may be generated using the speed prediction. The vehicle crash prediction may then be transmitted to a remote device.