Particle Filter Based Algorithm for Target Position Estimation Under Sparce Sensor Surveillance

A particle filter based algorithm was developed to track vehicles in a network of roads under the assumption of sporadic and non-persistent sensor data. It is assumed we have a number of sensors that provide position and velocity information only, which are scattered at possibly uneven intervals throughout the road system of interest. Further, the sensor ranges do not overlap, meaning we do not have constant eyes on target. The algorithm was based on the particle filter, but differed from the classical particle filter in two fundamental ways. First, particle weights are not used. Instead, a correspondence function is calculated only when a sensor is tripped, giving weight to the validity of the sensor report. Potentially this results in a computational savings. Second, we do not periodically resample particles. Results demonstrate the approach can effectively track multiple targets in simulations with sparse surveillance