Stochastic complexity integral image based technique for fast video tracking

We propose a new method based on the minimization of the stochastic complexity for fast and efficient tracking adapted to video images with a static camera. The obtained criterion combines the advantages of background-subtraction-based techniques and those of using measures of similarities to a target model without requiring any tuning of a weighting parameter. It is then demonstrated that this approach can be implemented with a fast integral image technique to estimate the location and the rectangular shape of the target in a few milliseconds.