Bayesian Inference Approach to Learning Coordinated Traffic Behavior for Non-Tracking Sensors
Lawrence A. M. Bush, Dennis Ehn, Fang Liu · AIAA Infotech@Aerospace Conference · 2009
We have developed a method for learning and detecting coordinated traffic behavior for situations where tracking is infeasible. For example, tracking in large areas with dense traffic is very demanding of sensor resources. Therefore, we have explored non-tracking methods for interpreting intentional vehicle coordination. Our research is applied to wide area persistent surveillance using Moving Target Indication (MTI) radar data; a radar data processing technique for detecting moving vehicles. Our objective is to detect coordinated multi-site activity. Our approach is to statistically model traffic behavior classes, at an individual location, in order to detect location-level activity of interest. We then combine the results from multiple locations using a reasoning structure. We present a method for learning this reasoning structure from the data using a Bayesian network structure search. We then use the Bayesian network to infer the overall situation. We tested our approach by collecting MTI data while running multiple military ground scenarios, each involving coordinated activity over multiple sites. In our tests, we were able to reliably identify the overall behavior classes, providing strong support for our approach.