A Real Time Robust Hand Tracking Method with Normal Cameras
Driven by the wide usage of smart devices today, gesture control with normal cameras is in great need to be studied further, especially in healthcare sector, for example, patient monitoring and rehabilitation. This paper proposes a practical hand tracking
- PDF / 1,460,988 Bytes
- 10 Pages / 439.37 x 666.14 pts Page_size
- 17 Downloads / 202 Views
Hong Kong Applied Science and Technology Research Institute Company Limited HK, and Huawei Technologies Co., Ltd., Hong Kong, China [email protected] 2 Huawei Technologies Co., Ltd., Shenzhen, China
Abstract. Driven by the wide usage of smart devices today, gesture control with normal cameras is in great need to be studied further, especially in healthcare sector, for example, patient monitoring and rehabilitation. This paper proposes a practical hand tracking method to tackle the related challenges including efficiency and robustness. Based on an efficient framework which integrates segmentation and tracking, several enhancements are proposed for a robust hand chasing. Experiments on both PC and android smartphones prove the proposed method is efficient and robust. Keywords: Tracking · Hand · Robust · Real time · Normal camera
1
Introduction
Recent solutions to hand tracking have been using cameras with depth sensor which can produce a 3D point presentation of the hand [1] [2] [9]. Although these methods appeared attractive in some applications, for example, healthcare and character animation, the imaging hardware required is not often available for normal cameras on various devices, especially on the smart devices. On the other hand, gestures controls are already important to the devices with normal cameras. For example, gestures may be used in the smart phone navigation while the user is moving, or for a demo in a smart TV in an exhibition, or to simply control a PC with a normal camera in a presentation, or for a patient to tell a nurse that he or she has a special need. There is still a huge market for a real time (in PC over 30fps, in smart phone over 15fps) and robust hand tracking methods to support all the scenarios. Unfortunately, a normal camera system for hand tracking is often suffered from various challenges such as hand appearance changes, camera limitations, and the disturbing environments. In the related previous work, TLD proposed by Kalal [10] is a solid integration of detection and tracking but do not handle articulated objects well. Liu and his team proposed to combine skin color information to trace the hand in TLD but the tracking rate is hard to meet the real time requirements [3]. Zhang’s work for structure-preserving object tracker revealed improvements in multi-object © Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part I, CCIS 546, pp. 56–65, 2015. DOI: 10.1007/978-3-662-48558-3_6
A Real Time Robust Hand Tracking Method with Normal Cameras
57
tracking while the articulated objects still remained an issue. In addition, seldom previous work took the tracking failure into consideration to make the tracking as continuous as possible [1]-[3]. In our practice, we realized a single method may not tackle the challenges well and therefore multiple cues should be used to extract the hand in each frame. Accordingly, we tried to integrate segmentation and tracking first to have an efficient hand tracking scheme. After that, we enhanced tracking and segmentation separa
Data Loading...
