Multi-object tracking in video using Localized Generalization Error model based RBFNN

Objects in video are high level features which provide plenteous information for video analysis like video indexing, retrieval and understanding. In this paper, a new method of multi-object tracking in video has been proposed by combining Gaussian background modeling, Background subtraction, mean-shift algorithm and used Radial Basis Function Neural Network (RBFNN) optimized by the Localized Generalization Error to recognize and classify object candidates. Experimental results demonstrate that the proposed method yields a high accuracy in object tracking and less false tracking.