This paper proposes a low-complexity design for 3D hand tracking, which can provide depth information and is able to work under critical backgrounds. This paper also proposes an effective way to segment hands out of entire image and also facilitates depth estimation of tracked hands in real-time by dualcamera systems. Multithreading and several techniques are applied to reduce computational complexity in proposed design. The final algorithm has been implemented both on PCs (with Intel Core i7 processor) and an embedded system (with ARM Cortex A9 processor). On PCs, it reaches 24 frames per second at VGA video. On the other hand, after reducing image size (i.e. QVGA video), it achieves the performance about 8 frames per second on PandaBoard embedded system.