Integrating Differential Evolution and Condensation Algorithms for License Plate Tracking

In this paper, we propose a novel method for position estimation and tracking of license plates in 3D from monocular camera view. Given an initial estimate, we try to track the position of the license plate in the successive video frames. In our previous works, we utilized condensation algorithm for estimating the state of the object and filtering the measurements. In this work, we present DE-condensation algorithm, which is an integration of condensation and differential evolution (DE) algorithms. The new algorithm is composed of the probability density propagation of condensation algorithm and a fine scale optimization step according to DE algorithm. Each sample of the DE-condensation algorithm is an estimate for the license plate´s position in 3D and projected to the image plane by perspective camera model for evaluation. While tracking the license plate´s texture by probability density propagation, DE optimization gives a fine tuning according to the license plate boundaries. DE-condensation algorithm outperforms the standard condensation algorithm and reduces the number of samples significantly