Opto-electronic imaging and tracking system must be guaranteed to recognize a variety of targets. Meanwhile, the scheduled image processing is a hard real-time processing course. Our studies have developed a real-time tracking fusion approach including pixel-base, feature-based and decision-based fusion, which simultaneously run multiple algorithms suiting to independently analyze and obtain various feature of targets for satisfying different targets in different tasks. A prototype has been developed to adaptively track motive objects. The paper introduces in detail the idea and method of real-time tracking fusion technology, and specifies hardware design and algorithm performance by analyzing embedded parallel architecture. The reliability, stability, adaptivity and real-time performance of the approach have been widely proven in real applications.