Neural network for object tracking

Abstract

Real-time object tracking is a problem which involves extraction and processing of critical information from complex and uncertain image data in a very short time. In this study, we present a global-based approach for object tracking in video images. Knowing grey level difference between target and estimated region containing the tracked object, we employ an Artificial Neural Network (ANN) to evaluate the corrective vector which is used to find the actual position of the target. Before, this ANN has been trained, during an offline stage, over a set of output and input samples to determine the relation between the intensity variations and position variations. The evaluation of the corrective vector can be obtained with small online computation and makes real-time implementation on standard workstations possible. © 2007 Asian Network for Scientific Information.