基于深度传感器的手势追踪系统设计与实现

发布时间：2018-09-18 20:00

【摘要】：机器视觉、人机交互如何更好地服务于人类是当今世界的主流研发问题。伴随着深度传感器技术成熟,机器视觉得以更加广泛地应用于人类的学习生活中,类似于扫地机器人等智能家居就是人工智能深入到人类生活的体现。其中手势是一种非常方便高效的交互手段,使用辅助传感设备例如数据手套、体感技术等的研发不断拓宽了人机交互的渠道,其中包括对于人手的识别和追踪的研究。本文对于手势追踪的研究与实现,主要是想将此功能应用在家庭环境中,结合智能家居等设备,给家庭生活中行动不便的老人或者残身体有所缺陷的人提供帮助,使手势追踪这一功能具体呈现在让家庭生活中,更好地服务于人类社会。为了实现裸手追踪系统,本文主要使用了粒子滤波算法,是对马尔科夫-蒙特卡洛算法的一种改进。不同于普通的追踪,人体手部颜色相对单一,形状在不同时刻有所变化,故本文将颜色特征和深度信息相互关联起来使用,结合人体裸手的形状变化以及颜色亮度特征,基于深度传感器对人手进行追踪。本系统使用上一时刻的手势运动速度、位置、形状、目标所占区域大小等信息,对当前时刻的手势位置进行了预测。在预测的计算过程中使用了积分图,使得前后时刻的运动状态相差减少,更加有效地使用上一时刻的信息预测当前手势的位置。本文对系统所实现的手势追踪在不同光照、不同手势状态、遮挡等情况下进行了测试,由实验结果可知光照情况对手势识别追踪有一定的影响,通过对假设目标区域的空洞填补克服了光照的影响,手势在相邻时刻发生变化、被遮挡的情况下通过积分图提高效率,可以实现相邻时刻变化差异可追踪,使上一时刻追踪到的目标特征信息应用到当前时刻的追踪中,根据目标距离变化、区域所占像素点个数差异、速度变化等信息预测当前时刻的目标位置,使追踪结果更加可靠。故本系统可以在不同状态下实现追踪,效率较高,结合在日常生活中用到的手势,从而可以根据手部运动速度的缓急方向、移动的轨迹分为多种不同的含义,将此系统应用于家庭服务机器人上,使人机交互更加便捷。
[Abstract]:Machine vision and human-computer interaction are the main research and development problems in the world. With the maturity of depth sensor technology, machine vision can be more widely used in human learning life. Intelligent home such as floor sweeping robot is the embodiment of artificial intelligence in human life. Gesture is a very convenient and efficient means of interaction. The research and development of auxiliary sensing devices such as data gloves and somatosensory technology has continuously widened the channels of human-computer interaction, including the research on human hand identification and tracking. The purpose of this paper is to apply this function to the home environment, combine with the intelligent home and other devices, to provide help to the old people who are physically disabled or the disabled people in their family life. The function of gesture tracking is presented in the family life and serves the human society better. In order to realize the naked hand tracking system, the particle filter algorithm is mainly used in this paper, which is an improvement to the Markov Monte Carlo algorithm. Different from ordinary tracking, the human hand color is relatively single, and the shape changes at different times. Therefore, the color feature and depth information are used in this paper, combined with the shape change and color luminance feature of human bare hand. The human hand is tracked based on the depth sensor. The system uses the information of the motion speed, position, shape and the size of the target area to predict the gesture position at the current moment. The integral graph is used in the calculation of the prediction, which reduces the difference between the motion state of the time before and after, and more effectively uses the information of the previous moment to predict the position of the current gesture. In this paper, hand gesture tracking is tested under different illumination, different gesture status, occlusion and so on. From the experimental results, it can be seen that illumination has a certain influence on gesture recognition tracking. The effect of illumination is overcome by filling the holes in the hypothetical target area, the gesture changes at the adjacent moment, and the integral image improves the efficiency in the case of occlusion, and the difference between the adjacent moments can be traced. The target feature information which was traced at the last moment is applied to the tracking of the current time, and the target position at the current time is predicted according to the target distance change, the difference in the number of pixels occupied by the region, and the change of the velocity, so that the tracking results are more reliable. Therefore, the system can be tracked in different states, high efficiency, combined with the hand gestures used in daily life, thus can be divided into a variety of different meanings according to the priority direction of the hand movement speed. This system is applied to home service robot, which makes man-machine interaction more convenient.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP212

【参考文献】