In this chapter we describe the particle estimators and its effectiveness for tracking objects in video sequences. The particles estimators are specifically advantageous in transition state models and measurements, especially when these are non-linear and not Gaussian. Once the target object to follow has been identified (in position and size) its main characteristics are obtained using algorithms such as FAST, SURF, BRIEF or ORB. As the particle estimator is a recursive Bayesian estimator, where observations update the probability of validating a hypothesis, that is, they use all the available information to reduce the amount of uncertainty present in an inference or decision problem. Therefore, the main characteristics of the object to follow are those that will determine the probability of validating the hypothesis in the particle estimator. Finally, as an example, the application of a particle estimator is described in a real case of tracking an object in a sequence of infrared images.
Part of the book: Digital Image Processing Applications