人体基础运动条件下的动态手势识别研究

发布时间：2018-06-29 00:25

本文选题：人体基础运动 + 动态手势识别　；参考：《电子科技大学》2014年硕士论文

【摘要】：随着科技的进步,移动电子设备上搭载了MEMS陨性传感器等丰富的硬件设施,给人们的生活带来了更多的便利,也给语音识别、图像识别、手势识别等新型人机交互方式提供了良好的平台。图像识别交互易受光线的影响、语音识别交互易被杂音干扰,于是,基于惯性传感器的手势交互凭借其独特的优势,成为当前研究的热点。人体静止情况下的动态手势识别有了很大的进展,许多学者尝试了不同的识别方法,并验证了其有效性。然而,现阶段对于人体处于运动状态下的手势识别研究几乎没有,本文将从这个方向入手,展开对人体运动条件下的动态手势识别研究。本文利用诺基亚公司提供的惯性测量装置Sensor-Box,重点研究步行、上下楼梯、电梯等人体基础运动条件下的动态手势识别。本文对人体基础运动进行了分类,并分析了人体基础运动对动态手势的影响。结合课题研究目标,提出了三种人体运动条件下的动态手势识别方案。从剔除人体基础运动干扰角度,提出了基于双惯性传感器法和数学模型法的识别方案；从特征分类识别角度,提出了基于阈值增大法的识别方案。为了减小惯性传感器的误差干扰,本文建立了Sensor-Box加速度计和陀螺仪的误差模型,并采用六位置法来标定其确定性误差。对加速度随机误差信号,采用时间序列分析法建立了ARMA模型,并用经典卡尔曼滤波器对随机误差进行了有效滤除。对于陀螺仪随机误差信号,采用标准Allan方差法分析识别其主要随机误差项,并采用小波分析对其噪声信号进行了有效分离。双惯性传感器法通过两个Sensor-Box同时采集人体运动信号和动态手势信号,然后基于相对运动理论来剔除人体基础运动的干扰,试验效果满足要求。数学模型法,针对不同类型的基础运动,提出了不同的数学模型,并给出了模型构造方法。对于人体步行、上下楼梯等周期性明显的信号,提出了一种建立周期信号数学模型的方法；该方法试验效果符合要求,算法还需完善。阈值增大法基于特征分类的方法,找出了翻转、甩动、晃动等动态手势的信号分类特征,建立分类器,试验效果良好。
[Abstract]:With the development of science and technology, mobile electronic devices are equipped with MEMS meteorite sensors and other rich hardware facilities, which bring more convenience to people's life, and also give voice recognition and image recognition. Gesture recognition and other new human-computer interaction methods provide a good platform. Image recognition interaction is easy to be affected by light, and speech recognition interaction is easily interfered by noise. Therefore, gesture interaction based on inertial sensor has become the focus of current research because of its unique advantages. There has been great progress in dynamic gesture recognition in static human body. Many scholars have tried different recognition methods and verified their effectiveness. However, there is almost no research on gesture recognition under the condition of human motion at the present stage. This paper will start with this direction and carry out the research on dynamic gesture recognition under the condition of human motion. In this paper, the sensor Box, an inertial measuring device provided by Nokia, is used to study the dynamic gesture recognition under the basic motion conditions of human body, such as walking, up and down stairs, elevators and so on. In this paper, the basic motion of human body is classified, and the influence of basic motion on dynamic gesture is analyzed. Combined with the research object, three dynamic gesture recognition schemes under human motion conditions are proposed. From the angle of eliminating human basic motion interference, this paper proposes a recognition scheme based on double inertial sensor method and mathematical model method, and from the point of view of feature classification recognition, puts forward a recognition scheme based on threshold enlargement method. In order to reduce the error interference of inertial sensor, the error model of sensor-box accelerometer and gyroscope is established in this paper, and its deterministic error is calibrated by the six-position method. The ARMA model of acceleration random error signal is established by time series analysis and the classical Kalman filter is used to filter the random error effectively. For the gyro random error signal, the main random error terms are identified by standard Allan variance method, and the noise signal is effectively separated by wavelet analysis. The dual inertial sensor method collects the human motion signal and the dynamic gesture signal simultaneously by two sensor boxes. Then based on the relative motion theory the interference of human basic motion is eliminated. The experimental results meet the requirements. In this paper, different mathematical models are proposed for different types of basic motion, and the method of model construction is given. For the periodic signals such as walking and going up and down stairs, this paper presents a mathematical model of periodic signals, the experimental results of which meet the requirements, and the algorithm needs to be improved. Based on the feature classification method, the threshold enlargement method finds out the signal classification features of the dynamic gestures such as flipping, shaking and shaking, and establishes the classifier, and the experimental results are good.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP212;TN911.7

【相似文献】