(An) active contour model for object tracking using prior contour

Abstract

Active contour models for object tracking using prior contour are investigated. The proposed algorithm uses edge or color feature for tracking and also uses the prior contour information for robust characteristics. The orthogonal search to contour extend search range and reduce computational burden compared with conventional snake algorithms, Kass``s variational method, Amini``s dynamic programming, William``s greedy algorithm. The conventional algorithms has many problem in tracking, so model techniques are noticed in tracking with snake. The tracking with models such as deformable model, eigenvector model and dynamic model are suggested. But they need so complex procedure such as training, estimation and matching. The pro-posed algorithm suggests the tracking scheme such like Kass``s one, energy minimization problem. But it use the prior contour as a model defining smooth similarities with two contours. The algorithm search normally and use dynamic programming to solve energy minimization problem. The main drawbacks of conventional snake algorithm - shrinking, limited search range, sensitive to noises - improved with the proposed algorithm. The external energy using color is also introduced. This color information can get from prior contour. It is updated for each frame and give the robust characteristic to track a solid-color objects. It can be adopted to robocup to detect ball or patch-tracking application because of the ability to be able to use shape and color information in the same time.