(I) Technologies to be developed

• Feature Tracking


To obtain motion information, features of objects need to be extracted, and then tracked to establish their correspondence across multiple video frames. The most commonly used feature is corner [17,18,20]. However, it is not robust enough in most of the cases as shadow or reflection on object surfaces are usually detected as corners, which are not reliable and could introduce noises in establishing feature correspondence. We aim to come up with a robust feature extraction method to enhance matching rate and accuracy of feature correspondences.


• Parameterized motion model Fitting


With calibrated camera parameters and corresponded features, we intend to develop a method to fit the aforesaid parameterized multiple motion model, and derive instantaneous motion of each pixel on the object for use by motion deblurring process.

(II) Innovation use of existing technologies

• Camera Calibration


This technique is related to the estimation and modeling of every parameter of a given camera such as aperture, focus distance and geometrical features. An accurate camera calibration is pivotal to the success of motion information extraction. We will leverage our camera model in [19,21] to develop a calibration method based on a calibration pattern derived from typical geometrical features within the scene.


• Deblurring and Super Resolution


Once the motion of the object of interest is estimated, deblurring could be achieved by Wiener deconvolution, and existing super-resolution technologies could be adopted to fuse different snapshots of the same object of interests to result in an enhanced snapshot with better clarity.