Real-Time Hand Gesture Recognition Based on a Fusion Learning Method

With the development of artificial intelligence, hand gesture recognition plays a significantly important role in human-computer interaction applications. However, due to the complex background and volatile illumination, it is still challenging to recognize hand gestures in real-world scenarios. To overcome the mentioned problems and achieve real-time dynamic and accurate recognition, a novel fusion learning method is proposed based on depth images. The algorithm is composed of efficient statistical hand segmentation algorithm, static hand pose classification via simplified shallow CNN and an improved model matching method. The experimental results demonstrate that the proposed method shows remarkable performance in complex real-world scenarios. The accuracy of the proposed method is over 98% running on the CPU, 0.08s per frame. Compared with end-to-end DNN, our algorithm has lower computational complexity and better applicability.