Fast and Effective Visual Tracking Method for Multiple Moving Targets

摘要

In this paper we describe the tracker that provides the real-time visual feedback using on-board low-cost computers. Instead of using the voluminous and costly special purpose image processor, we exploited conventional processors and reduced processing time using the smart system software and vision software. The proposed tracker is based on the Two Stage Visual Tracking Method (TSVTM) which consists of real-time kernel, image saver, and vision module. A vision task is divided into two stages according to the information available: recognition and tracking itself. In the recognition stage, grayscale images are segmented into a set of regions. Targets are then recognized by their shapes. Once targets are initially located by the recognition stage, tracker itself can easily find and keep tracking targets using the focus-of-attention strategy based on Kalman filter. In order to avoid its inherent large variation of response time due to the difference of computation requirements between both stages, real-time kernel and image saver manage the resources efficiently. The tracker has been implemented on an mobile robot CAIR-2 and exhibited at '93 World Exposition and '95 IJCAI Robot Competition[19, 21].