Analysis and integration of road projection methods for multiple ground target tracking

This paper analyses and evaluates two projection families and describes their integration in a multi target tracker expected to be used in a real world experimentation. The knowledge of the road network allows for more reliable track maintenance and accurate state estimation, compared to standard trackers, when targets are closely spaced and undertake maneuvers The methods to be studied are the orthogonal projection, and the probabilistic projection, which are applied to the state estimate. The tracker's estimator uses directional dynamic models corresponding to the roads of interest. The use of competing non-interacting models instead of the interacting multiple model (IMM) method is explained and justified in the context of road targets. The multimodel estimator computes costs, transmitted to an S-dimensional assignment algorithm which determines the best measurement-to-track association. Simulation results show the effectiveness of each projection method on the overall tracking precision, in specific road network and sensor configurations.