3D tracking of license plate from monocular camera view

In this paper, we present a method for tracking a planar object (license plate) in 3D. Given an initial estimate, we try to estimate the location, motion vector and pose of the object in 3D for the successive video frames. We utilize condensation algorithm for estimating the state of the object and filtering the measurements according to the extracted image features. Image features, namely the directional gradients on object boundaries, are calculated locally for each guess of the condensation algorithm. Local extraction of the features decreases the computational complexity of the tracking process. The algorithm tries to fit the state of the object to the image gradients. In contrast with optimization-based approaches, condensation algorithm simplifies the complexity of the system. The parameters of the algorithm are easy to understand and easy to apply even for complex structures. We demonstrate the results of the method for tracking a license plate for different scenarios. The results are promising for further research on the method.