Milestones

To achieve the first milestone, we shall implement the camera calibration and feature correspondence module based on our previous researches on calibration and corner detection proposed in [18, 19, 21]]. Lane markings will be automatically detected as calibration pattern, and then extrinsic parameters of camera will be estimated. On the other hand, feature points which are robust for tracking will be extracted by improved corner detection algorithm.

With estimated camera parameter, we can calculate the image coordinates offset of any feature point with forward motion and/or rotating motion. As its inverse process, with sufficient tracked feature points and the assumption that the target object is rigid, we can fit the proposed multiple motion model onto each detected object based on iterative optimization or any error minimization approach, so that the rotating motion and forward motion can be accurately estimated to achieve our second milestone.

The key to the success of motion deblurring technology is to predict accurately the motion of objects. With correctly estimated motion, we can then realize deblurring based on Wiener Deconvolution. Estimating motion and deblurring of objects on a per-object basis in a scene can provide more accurate result than those obtained by processing the entire scene as a single object. On the other hand, the feature correspondence module also eases the task of image registration, which can be utilized to get super-resolution images based on the knowledge in our previous ITF project (ITS/174/08).

Upon the completion of the above three milestone, we will then implement a prototype forensic video enhancement application to demonstrate the robustness and effectiveness of the technologies developed in this project.