基于Kinect的人机交互技术及在矿井火灾逃生模拟系统中的应用

发布时间：2018-06-22 22:44

本文选题：虚拟煤矿火灾 + Kinect人机交互　；参考：《山东科技大学》2017年硕士论文

【摘要】：随着虚拟现实和Kinect人机交互技术的发展,Kinect越来越多的被应用到体感游戏及大型场景展示上,用来与虚拟场景进行互动。本文主要研究了基于Kinect的人机交互技术及其在虚拟煤矿巷道火灾逃生模拟系统中的应用。首先研究了整个系统的软硬件配置,其次研究了如何利用Kinect捕捉人体深度图像以及骨骼数据;然后研究了基于深度图像的正反手势识别,以及基于骨骼数据的体态识别;最后研究了利用UE4创建虚拟巷道模型,用正反手势控制巷道中人物的前后左右行走以及利用步态识别算法识别操作者的身份,并将该操作者的信息在虚拟巷道中显示出来。在进行正反手势识别时,首先利用Kinect for Window SDK 2.0的库函数读取深度信息,再利用阈值分割将手部区域分割出来,然后将手部图像分为两层:第一层为直立手指层,第二层为蜷缩手指层。对于直立手指层,先利用轮廓检测算法检测出手部轮廓,再利用凸包检测算法识别出手指的个数;对于蜷缩手指层,主要利用轮廓检测算法检测轮廓,有轮廓则为正手,无轮廓则为负手。最后把这两层识别结果结合起来,实现正反手识别。步态识别也是先通过Kinect体感设备及Kinect for Windows SDK 2.0读取骨骼数据。进行识别之前,首先分析人走路的姿态,提取特征向量。人走路时主要特征有速度,步长,各个肢体摆动的幅度,本文中主要选择10个肢体角度进行分析。然后,画出每个肢体角度随时间变化的曲线,从中提取一个平均周期,并用多项式函数拟合。最后,用多项式的拟合参数作为特征,用KNN分类算法,与数据库中的参数进行匹配,识别出人的身份。
[Abstract]:With the development of virtual reality and Kinect human-computer interaction technology Kinect is more and more used in body sense games and large-scale scene display to interact with virtual scene. This paper mainly studies the human-computer interaction technology based on Kinect and its application in virtual mine tunnel fire escape simulation system. Firstly, the hardware and software configuration of the whole system is studied, secondly, how to capture the human depth image and bone data by Kinect, then the forward and negative gesture recognition based on the depth image and the posture recognition based on the bone data are studied. Finally, the virtual laneway model is created by using UE4, the forward and left walking of the characters in the roadway is controlled by positive and negative gestures and the identity of the operator is recognized by gait recognition algorithm, and the information of the operator is displayed in the virtual laneway. When using the library function of Kinect for window SDK 2.0 to read the depth information, the hand region is segmented by threshold segmentation, and then the hand image is divided into two layers: the first layer is the vertical finger layer. The second layer is the crouching finger layer. For the vertical finger layer, the contour detection algorithm is first used to detect the contour of the hand, and then the number of fingers is recognized by using the convex hull detection algorithm; for the curled finger layer, the contour detection algorithm is mainly used to detect the contour, while the contour is forehand. No contours are negative hands. Finally, the recognition results of the two layers are combined to realize forward and backhand recognition. Gait recognition also uses Kinect somatosensory devices and Kinect for SDK 2.0 to read bone data. Before recognition, the attitude of human walking is analyzed, and the feature vector is extracted. The main characteristics of human walking are speed, step size and swing amplitude of each limb. In this paper, 10 limb angles are selected for analysis. Then, the curve of each limb angle changing with time is drawn, from which an average period is extracted and fitted with polynomial function. Finally, the fitting parameters of the polynomial are used as the feature, and the KNN classification algorithm is used to match the parameters in the database to identify the identity of the person.
【学位授予单位】：山东科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD752;TP391.41

【参考文献】