No abstract

No abstract

No abstract

INTRODUCTIONWith development of image processing and machine vision, more flexible and convenient human-machine interaction has been a hot-spot in the field. As one of crucial technologies, hand gesture recognition attracts many researchers' attention. Up to now, there are about two types of features which are applied for hand gesture recognition, global features and local features. Fourier descriptors[1], Hu moments[2]and features of spatial distribution[3] are belonged to global features, but which require precise skin segmentation to reserve hand gestures' true shapes. However, skin detection is a difficult task in real environment. Uneven illumination, blurred images and skin color objects in backgrounds lead to poor skin segmentation. Local features include SIFT[4], SURF[5], HOG[6], LBP[7] descriptors and so on. SIFT and SURF can't eliminate feature points in background, and their calculations are very complex. LBP is one of texture descriptors, but it is sensitive to noises. Paper[8] compared performance of Zernike, HOG and SIFT as hand gesture features in cluttered background, results indicate that HOG can achieve the highest recognition rate, which is more suitable for practical application. HOG can represent local appearance and shape of objects well[9], moreover, it is not sensitive to non-uniform light and small-angle rotation, so it is widely used in fields such as pedestrian detection, face recognition and hand gesture recognition. Different improved HOG features are proposed in different application. Pyramid histograms of oriented gradients [10] are proposed by Anna Bosch in 2007. An image in each layer is divided into several equal-sized cell and sizes of cells in different layer can be different. Then gradients in different layers are concatenated to form final feature. By this way, the method can depict more comprehensive characteristics about local shape of the objects. Because HOG is a representation of edges, both gradients of hand gesture and background are included in the features. Paper [11] introduced a fixed value as the weight of gradient magnitude when the pixel is identified as skin color. To some degree, this method can strengthen features of skin color region, but it is difficult to decide precisely whether a pixel is skin, in addition, how to choose a suitable value to strengthen skin information is an open research question. A new method to weaken non-skin pixels' gradients information is proposed in this paper. It needs not to determine whether a pixel belongs to skin color. A skin similarity [12] can evaluate importance about gradient magnitude of the pixel so that it is introduced to be as the weight of the gradient magnitude. Furthermore, HOG of two sized cells are concatenated as final hand gesture features. SKIN COLOR BLOCKS LOCATIONSkin color detection is one of most common methods to locate hands in whole images. There are some methods such as simple threshold model, single Gaussian model, mixed Gaussian model and so on to be used to identify the skin pixels in an image. 12...

Data fusion in the Internet of Things (IoT) environment demands collecting and processing a wide variety of data with mixed time characteristics, both real-time and non-real-time data. Most of the previous research on data fusion was about the data processing aspect; however, successful data transmission is a prerequisite for high-performance data fusion in IoT. On the other hand, research on data transmissions in IoT mainly focuses on networking without sufficiently considering the special requirements of the upper-layer applications, such as the data fusion process, that are consuming the transmitted data. In this paper, we tackle the problem of data transmission for data fusion in an IoT environment by proposing a distributed scheduling mechanism VD-CSMA in wireless sensor networks, which considers the values for data fusion, as well as the delay constraints of packets when determining their priority levels for transmission. Simulation results have shown that VD-CSMA may enhance both throughput and delay performance of data transmission as compared to the typical scheduling schemes used for data fusion in IoT.