JIITA, vol. 2, no. 3, pp.140-144, Sep, 2018
Depth Noise Filtering Method based on Quadrangular Cutoff Region and Dynamic Operation Bound for Hand Gesture Recognition
Yoon-Jib Kim *, Sang-Heon Lee *, Byunghun Hwang, Hyunduk Kim, Youngsun Ahn
Daegu Gyeongbuk Institute of Science & Technology (DGIST), Daegu, Republic of Korea
Abstract: In this work, we propose how to remove unwanted noise from incorrect information from a depth camera used for a hand-gesture recognition system. When detecting a hand, the machine is very likely to fail to catch the closest point on the subject to its sensor due to the low resolution of the depth sensor or to the rough surface of a hand. Thus, we introduce a novel approach to remove such noise effectively using a Quadrangular Cutoff Region. Moreover, in order for a computer to recognize manual gestures, it needs several operation boundaries to notify the machine that a hand is coming. Therefore, the positions of the boundary areas need to be fitted to the stature of people so that the system can discern hand movement correctly in comparison with other movements. This work proposes a Dynamic Operation Bound that changes its position vertically according to the height of human users.
Keyword: OpenNI; noise filtering; hand detection; gesture recognition; depth camera;