| 题名 | 基于Kinect的人机体感交互关键技术研究; Research on Key Technologies of Human-Computer Motion Sensing Interaction Based on the Kinect |
| 作者 | 何晓鹏 |
| 答辩日期 | 2014 ; 2013 |
| 导师 | 吴清锋 |
| 关键词 | 体感交互 姿势校正 手势识别 Motion Sensing Interaction Posture Correction Gesture Recognition |
| 英文摘要 | 近年来，随着运动捕捉技术的发展并逐步成熟、Kinect低价体感设备的面市及其开发工具KinectSDK的发布，运用廉价的Kinect设备进行人机体感交互的开发逐渐成为研究的热点，目前在体感游戏、网络社交、购物、医疗领域中已出现基于Kinect的人机体感交互应用。但由于基于Kinect的人机体感交互抛弃设备而使用影像辨识，使得Kinect会由于获取的运动信息不平滑，导致交互不能够被准确识别。同时在进行动作捕捉时不借助设备，无法准确定位人体关节的位置，导致程序进行动作、姿势识别时，对部分动作、姿势无法准确识别。因此如何优化Kinect生成的原始骨骼信息，使Kinect获得的骨骼信息准确、运动信息平滑，以提高手势识别与姿势识别的精度，对于基于Kinect的人机交互具有十分重要的意义。本文围绕基于Kinect的人机体感交互的两个识别方式：手势识别与姿势识别，对其中影响交互效果的运动信息不平滑、姿势定位不准确等问题展开深入研究，主要工作包括： （1）研究基于DTW的手势识别优化。本文在研究DTW手势识别的基础上，对进行识别的手势进行运动平滑，消除骨骼运动的抖动，在视觉上获得较平滑的运动效果，并使得DTW算法适应于Kinet的手势识别。首先，分析了限制DTW算法应用于Kinect手势识别的主要原因：骨骼运动的信息抖动；其次，分析了SDK中自带的骨骼运动平滑方法及其不足；在此基础上对SDK中的骨骼运动平滑算法进行改进，提出主要参数能够依据关节运动快慢调整的方法并以此对DTW手势识别方法进行改进，提高手势识别精度。实验表明，本文的改进的DTW手势识别方法可以运用于Kinect手势识别，对实验手势有着更高的识别率。 （2）研究基于长度与坐标约束的姿势优化。本文研究了关节被遮挡姿势的识别优化标准流程，借鉴其他骨骼模型的校正方法，使其适用于Kinect生成的姿势校正。首先，根据关键节点确定查找被遮挡关节的遍历路径；其次，应用关节的深度值在前景、背景的数值不同，确定被遮挡关节；然后，利用被遮挡关节在模板姿势下的距离校正被遮挡关节的长度，根据被遮挡关节与前驱、后继关节构成的三角关系校正被遮挡关节的坐标。实验结果表明，本文的方法能够使关节被遮挡情况下的姿势识别效果更准确。; In recent years, with the development of motion capture technology, Kienct motion sensing interaction equipment marketed and its developing tools Kinect SDK release, the use of cheap Kinect device for the development of the humancomputer motion sensing interaction has become a hot research. Currently in motion sensing games, social networking, shopping, and medical fields have emerged a sense of motion sensing applications based on Kinect. However, due to Kinect abandoned equipment, using image recognition, making the motion information acquired by Kinect is not smooth, interaction cannot be accurately identified. In addition, the motion capture equipment is not used, it can not accurately locate the position of human joints and cause on the part of movements, also, gestures can not accurately identify. How to optimize the original skeletal information generated by the Kinect, improved information effect, keep the skeletal information and motion information obtained by Kinect accurate have a significance of the Kinect-based human-computer interaction. This dissertation focuses on two identification methods of motion sensing interaction of Kinect: gesture recognition and pose recognition, and research two key issues that affect the interaction effect: the Smooth of the motion information and the Joint blocked posture recognition, the main work includes: (1) Research on the optimization of DTW-based gesture recognition. In this dissertation, on the basis of the study of DTW gesture recognition, smoothing the gesture movement, to improve the recognition rate of the gesture. Firstly, analyze the main reasons impact the effect of gesture recognition: jitter of Joint movement. Secondly, analyze the smoothing method in the SDK and its shortcomings; on this basis, improving smoothing algorithm in the SDK and making the main parameters of smoothing algorithm can be adjusted based on the movement speed, to improve the precision of gesture identification. Experimental results show that the improved method in this dissertation has a higher recognition rate than the traditional DTW gesture recognition method. (2) Research the optimization of posture recognition by length and coordinate constraint. This dissertation research the optimization process of joint blocked posture recognition, and analyze the algorithm use in each process. Firstly, determine the key node traversal path of the block joint. Secondly, use the value of depth map different in the foreground and background to determine the blocked joints. Thirdly, use the length of the blocked joints in the template to correct the length of the joint; use the triangular relationship made of blocked joint and the precursor, and the subsequent to correct the coordinates of blocked joint. The experimental results show that the method In this paper, enable more accurate results of pose recognition with blocked joints.; 学位：工程硕士; 院系专业：软件学院_软件工程; 学号：24320101152259 |
| 语种 | zh_CN |
| 出处 | http://210.34.4.13:8080/lunwen/detail.asp?serial=41349 |
| 内容类型 | 学位论文 |
| 源URL | [http://dspace.xmu.edu.cn/handle/2288/77778]  |
| 专题 | 软件学院－学位论文 |
| 推荐引用方式 GB/T 7714 | 何晓鹏. 基于Kinect的人机体感交互关键技术研究, Research on Key Technologies of Human-Computer Motion Sensing Interaction Based on the Kinect[D]. 2014, 2013. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
修改评论