[雙語(yǔ)翻譯]手勢(shì)識(shí)別外文翻譯--光照不變的實(shí)時(shí)魯棒手勢(shì)識(shí)別

1、4100 英文單詞， 英文單詞，20500 英文字符，中文 英文字符，中文 6600 字文獻(xiàn)出處： 文獻(xiàn)出處：Chaudhary A, Raheja J L. Light invariant real-time robust hand gesture recognition[J]. Optik, 2018, 159: 283-294.Light invariant real-time robust hand gesture recogn

2、itionAnkit Chaudhary, J.L. RahejaAbstractComputer vision has spread over different domains to facilitate difficult operations. It works as the artificial eye for many industrial applications to observe elements, process,

3、 automation and to find defects. Vision-based systems can also be applied to normal human life operations but changing light conditions is a big problem for these systems. Hand gesture recognition can be embedded with ma

4、ny existing interactive applications/games to make interaction natural and easy but changing illumination and non-uniform backgrounds make it very difficult to perform operations with good image segmentation. If a vision

5、 based system is installed in public domain, different people are supposed to work on the application.This paper demonstrates a light intensity invariant technique for hand gesture recognition which can be easily applied

6、 to other vision-based applications also. The technique has been tested on different people in different light conditions with the extreme change in intensity. This was done as one skin color looks different in changed l

7、ight intensity and different skin colors may look same in changed light intensity. The orientation histogram was used to identify unique features of a hand gesture and it was compared using super- vised ANN. The overall

8、accuracy of 92.86% is achieved in extreme light intensity changing environments.Keywords: Gesture recognition ；Orientation histogram；Light intensity invariant systems ；Extreme change in light intensity ；Natural computing

9、；Robust skin detection1. IntroductionComputer vision applications have been part of industry operations for more than four decades. They are helpful in fasting the industrial process, automated many difficult tasks and a

10、lso help in finding minor defects [1]. Many applications were using hand gesture recognition techniques for different purposes as hand gesture provides a natural way to communicate with machines [2–5]. These applications

11、 were initially based on wired gloves, color strips or chemicals to detect a region of interest (ROI) smoothly. A survey of different devices and techniques used for hand gesture recognition can be found out in [6]. To m

12、ake human-machine communication more effective, gesture recognition of bare hand had introduced where any person could use his hand in natural position [7–10]. A lot of work has been done in the area of natural hand gest

13、ure recognition to make it more robust. Currently, this kind of applications [11] and games are more popular as a user feel comfortable and don’t need anything to operate the vision-based system.Recently there has been a

14、 growing interest in the field of light intensity-invariant object recognition. For advanced applications in this area, one can set up a system in the laboratory with ideal conditions. However, in practical scenarios, th

15、e vision systems may have applications in public domain where the daylight can also have effect. The light intensity may not be the same everywhere, hence a robust system that operates in all types of light conditions is

16、 required. The light intensity has a large impact on the visual image processing algorithms as the image segmentation depends on the colors in the image and these colors depends on light intensity at the time of image ca

17、pturing. If we are working with human skin detection then the scenario is not letters or characters, biological cells, electronics waveforms or signals, states of a system or any other items that one may desire to classi

18、fy. Any pattern recognition system consists of two components, namely feature transformation and classifier. The observation vector is first transformed into another vector whose components are called features. These are

19、 intended to be fewer in numbers than the observations collected, but must collectively represent most of the information needed for the classification of patterns. By reducing the observations to a smaller number of fea

20、tures, one may design a decision rule which is more reliable. For a given number of training samples, one can generally obtain more accurate estimates of the class conditional density functions and thus formulate a more

21、reliable decision rule [22]. In the past several methods have been used for feature extraction. Generally, the features are suggested by the nature of the problem. In image processing applications, the light intensity pl

22、ays an important role because it significantly affects the segmentation of the ROI from the original image frame. If the light intensity changes, then the threshold for skin filter also has to be changed. This motivates

23、the development of techniques that are applicable to different light intensities.Orientation Histogram (OH) technique for feature extraction was developed by McConnell [23]. The major advantage of this technique is that

24、it is simple and robust to lighting changes [24]. If we follow pixel-intensities approach, certain problems arise due to varying illumination [16]. If pixel by pixel proximity for the same gesture is taken from two diffe

25、rent images, while the illumination conditions are different, the distance between them would be large. In such scenarios, the picture itself acts as a feature vector. The main motivation for using the orientation histog

26、ram is the requirement for lightning and position invariance. Another important aspect of the gesture recognition is that irrespective of the orientation of the hand in different images, for the same gesture we must get

27、the same output. This can be done by forming a local histogram for local orientations [25]. Hence, this approach must be robust for illumination changes and it must also offer translational invariance.We would also need

28、the gestures to be the same regardless of where they occur in the image. The pixel levels of the hand would vary considerably with respect to light, on the other hand, the orientation values remain fairly constant. We ne

29、ed to calculate the local orientation from the direction of the image gradient. The local orientation angle & will be a function of position x and y, and the image intensities I(x, y). The angle & is defined as:θ

30、(x, y) = arctan[I(x, y)-I(x-1, y), I(x, y)-I(x, y-1)](1)Now form a vector Ф of N elements, with the ith element showing the number of orientation elementsθ(x, y) between the angles [i – 1/2 ] and [i + 1/2 ]. Where Ф is

31、defined as:360°𝑁360°𝑁3. Light invariant systemThe hand gesture recognition system works on the principle of the 2D computer vision. The system has an interface with a small camera which captures

32、 users’ gestures. The input to the system is image frame of moving hand in front of a camera captured as a live video. The preprocessing of image frame was done as discussed in [26] with real-time constraint. The resulti

33、ng image would be the ROI, only hand gesture image. Now we do need to find out feature vectors from the input image to recognize it with the help of classifier.As this system was for research purpose only, we took only s