Image Target Location And Tracking Based On Affine Transform Model

Image target tracking is an important research topic in the field of computer vision. In an image sequence, the view of a target is not necessarily invariant. It often suffers such transformations as translation, rotation, scaling and other complex deformations. In this paper, we endeavor to carry out an efficient scheme, by which ideal location and tracking results can be obtained in the presence of target deformations. First, image matching and target tracking problems are investigated in a summarizing style. Some currently used image motion estimation techniques and their respective merits and demerits are analyzed. Then we discussed the corrlation-type image matching algorithms and translational motion model which customarily goes with the correlation-type algorithms, and a fast Zero mean Normalized Cross Correlation (ZNCC) algorithm is constructed based on the box filtering technique. After that, some basic theories related with the affine model-based tracking algorithm , such as image registration, image transformaton, affine transform, similarity measurement, "coarse-to-fine"strategy, Newton's iteration method and template update stratedy, are expatiated in detail. With the use of a robust error mapping function, an affine model-based target tracking algorithm is formulated, in which a quasi-Newton's iterative method is implemented in the optimization progress. Explicit and exhaustive interpretations are set forth concerning the theoretical origin, mathematical deduction, and complete programming details of the above-mentioned algorithm. Experimental results show that, the algorithm is reliable and efficient, and performs especially well under circumstances with target's affine deformations.