|
Abstract:
|
With the rise of autonomous of vehicles there is a growing need for the detection of these vehicles, especially in regards to national security. Research on machine learning in this field has been largely limited to the use from the human operator level of perception, using algorithms to predict and optimize future movement, resource management, and security. Machine learning can play a powerful role in detecting vehicles leveraging multi-dimensional information. In a multi-sensor operating mode unsupervised learning algorithms can be extremely powerful tools for detection. This paper explores how a collection of unsupervised learning algorithms, specifically highlighting clustering and anomaly detection can be used to improve detection capability. This paper uses a dataset from Air Force Research Lab that includes collections from seismic, acoustic, and RF sensors for a variety of vehicles. Using various clustering algorithms, we find that unsupervised learning models perform well in detecting when vehicles pass by close to the sensors. When using a combination of seismic and acoustic sensors, the models can pick up on the small movements with high accuracy.
|
