Overview of problems and techniques in target tracking

In many engineering applications, including surveillance, guidance, navigation, robotics, system control, or quality management, single stand-alone sensors or sensor networks are used for collecting information on time varying quantities of interest, such as kinematical parameters and classification attributes of moving objects (e.g. manoeuvring air targets, ground vehicles) or, in a more general framework, for tracking time varying signal parameters. More strictly speaking, in those applications the state of a stochastically driven dynamical system is to be estimated from a series of sensor data sets, also called scans or data frames, which are received at discrete instants of time (scan/frame time, revisit time, data innovation time). The individual output data produced by the sensor systems considered (reports, observations, returns, plots) typically result from complex estimation procedures themselves that characterise particular waveform parameters of the received signals (sensor signal processing). Provided the quantities of interest are related to moving point-source objects or small extended objects (radar targets, for instance), often relatively simple statistical models can be derived from basic physical laws, which describe their temporal behaviour and thus define the underlying dynamical system. The formulation of adequate dynamics models, however, may be a difficult task in certain applications.