Deformable Part Model Based Hand Detection against Complex Backgrounds
Hand detection is a challenging task in hand gesture recognition system and the detection results can be easily affected by changes in hand shapes, viewpoints, lightings or complex backgrounds. In order to detect and localize the human hands in static ima
- PDF / 2,278,343 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 44 Downloads / 216 Views
Abstract. Hand detection is a challenging task in hand gesture recognition system and the detection results can be easily affected by changes in hand shapes, viewpoints, lightings or complex backgrounds. In order to detect and localize the human hands in static images against complex backgrounds, a hand detection method based on a mixture of multi-scale deformable part models is proposed in this paper, which is trained discriminatively using latent SVM and consists of three components each defined by a root filter and three part filters. The hands are detected in a feature pyramid in which the features are variants of HOG descriptors. The experimental results show that the proposed method is invariant to small deformations of hand gestures and the mixture model has a good performance on NUS hand gesture dataset - II. Keywords: Hand detection Deformable part model features Complex backgrounds
Latent SVM
HOG
1 Introduction Hand gestures are important body languages in human daily communication. Traditional HCI (Human Computer Interaction) devices such as keyboard and mouse are subject to the limitations of operational distance and convenience, so it’s a natural way for us to interact with the computer using hand gestures. Hand gesture recognition has various applications such as sign language recognition, remote video conference, games and VR (Virtual Reality). Glove based and vision based methods are usually used in hand gesture recognition system, in which glove based method requires user to wear special gloves which can deliver the movements of hands and fingers to the computer [1]. Such an approach can accurately recognize various hand gestures in real time, but it is an unnatural and expensive way to interact with the computer because of the adoption of the complex glove equipment. Vision based hand gesture recognition has become popular in recent years, it doesn’t require the user to wear gloves and only a camera is used to capture images of hands, which is a natural and friendly way for us to interact with the computer [2]. Figure 1 shows the process of vision based hand gesture recognition.
© Springer Science+Business Media Singapore 2016 T. Tan et al. (Eds.): IGTA 2016, CCIS 634, pp. 149–159, 2016. DOI: 10.1007/978-981-10-2260-9_17
150
C. Zou et al.
Fig. 1. Process of vision based hand gesture recognition
There are two types of hand gesture used in HCI system, i.e. static gesture and dynamic gesture, in which static gesture positions remain unchanged during a period of time and dynamic hand gesture positions are temporal and change with respect to time [3]. Static hand gesture recognition becomes popular in recent years because dynamic hand gestures can be considered as actions composed of a series of static hand gestures. The most difficult problem of vision based static hand gesture recognition is to detect hands against complex backgrounds, although depth cameras such as Kinect, LeapMotion and RealSense are robust and precise to detect hands according the depth and image information, they ar
Data Loading...
