Measurement-to-track assignment methods in target tracking

Object tracking is a problem of obtaining usually positional and motional information about the object. The observation is usually in the form of a picture, radar or sonar image. A key element of the tracking problem is to estimate a (non-measurable) state of a stochastic system based on corrupted measurements. This problem is well treated in literature. If the objective is tracking multiple objects simultaneously or if a false detection appears due to a clutter, in addition it is necessary to solve the problem of measurement-to-track assignment. A usual procedure is to assign measurements to objects based on a prediction of their location. There are computationally cheap measurement-to-track assignment methods based on minimizing a statistical distance, which include the well-known strongest or nearest neighbor methods. On the other hand, there is the multiple hypothesis tracking method providing high quality of multiple target assignment. Unfortunately its computational costs are unbearable. The probabilistic data association method represents a compromise between quality and computational costs. The paper aims at introducing some measurement-to-track assignment methods. A special attention is paid to the measurement-to-track assignment methods of the nearest neighbor type. Its basic form based on minimizing the statistical distance with a single time step will be presented and also its variants considering multiple time steps simultaneously for the assignment will be introduced. The methods will be illustrated in a numerical example.