Long-range Hand Gesture Recognition with Joint SSD Network

Chengming Yi, Liguang Zhou, Zhixiang Wang, Zhenglong Sun, Changgeng Tan

Abstract: The hand gesture recognition plays an important role in the human-computer interaction (HCI) field. Previous works mainly focused on researching this task in the shorter distance, which can only be applied in a limited scenario due to the recognition distance constrains. In this paper, we propose a method, which is named Joint Single Shot Multibox Detector (JSSD) network, to solve the hard long-distance hand gesture recognition task. The method is based on the framework of SSD. Our model employed two SSDs for object detection. One is for the head-shoulder region detection and the other is for the hand gesture bounding box proposal and classification. We validate our proposed model on real data collected from USB camera at several distance levels. The experiment results show that our method is more robust and precise in recognizing the hand gestures at a longer distance. Concretely, JSSD network is able to recognize the hand gestures captured up to 6 meters away from the camera.