The specific object tracking algorithm based on Kalman filter in an environment where similar objects are existing

Abstract

Visual target tracking is one of the important issues in computer vision. Especially, the particular object tracking in an environment of multiple objects has many practical applications such as automated surveillance system and robot vision system. However, it is difficult to track a particular object because many problems are occurred in an environment of multiple objects such as merging and splitting. This paper proposes a particular object tracking algorithm when similar objects are present. This algorithm is based on the Kalman filter. But, standard Kalman filter requires one measurement for state estimation. However, if similar objects exist in the image, we obtain multiple measurements including true measurement and false measurements. So, we define new cost function to distinguish the true measurement from several measurements. Also, if merging and splitting problems are occurred, we deal with these situations by changing cost function. The computer simulation based on Matlab shows that the performance of the proposed algorithm is appropriate for tracking a particular object in an environment that similar objects exist.