面向人体运动捕获的姿态测量算法研究与实现

发布时间：2018-04-02 15:33

本文选题：运动捕获　切入点：椭球拟合　出处：《浙江大学》2017年硕士论文

【摘要】：体感交互是新一代人机交互的主要界面形式之一。为了实现人体运动和计算机的交互,首先要对人体的运动进行捕获识别。本文针对MEMS惯性式人体运动捕获系统中的传感信息获取和运动姿态分析两个环节,分别从姿态测量单元设计、传感器误差校正、姿态解算与姿态补偿融合等方面进行了深入研究。本文首先对人体运动捕获系统的整体框架进行了设计,结合系统需求设计了姿态测量单元来获取加速度、角速度和磁场分量等运动数据。考虑到系统的可扩展性和实时性,本文基于双缓存和时分复用机制设计了支持多个测量节点异步采集和同步聚合的通信协议。在分析了姿态测量单元中磁传感器受到的自身仪表误差和外界环境干扰之后,建立了多参数误差模型,并基于椭球拟合法对误差进行了有效补偿。同时在分析了陀螺仪中存在的随机漂移后,针对其难以直接量化处理的特点,构建了 ARMA模型,并使用卡尔曼滤波有效消除了陀螺仪的漂移。针对人体运动捕获系统中的运动姿态分析环节,本文在对陀螺仪积分和磁传感器倾斜补偿两种姿态解算方法进行实验对比的基础上,提出了多传感器补偿融合算法,即分别基于梯度下降法、基于互补滤波和PI调节实现了姿态融合,并在不同的状态下对融合精度进行了测试。测试表明,两种融合算法均达到了静态精度0.5°,动态精度2°。与另一常用的融合方法卡尔曼滤波相比较,此两种算法也能达到相应的精度。且在相同精度下,复杂度更小,更加适合于低功耗的嵌入式系统。
[Abstract]:Somatosensory interaction is one of the main interface forms of the new generation of human-computer interaction.In order to realize the interaction between human motion and computer, the first step is to capture and recognize human motion.In this paper, the sensor information acquisition and motion attitude analysis in the inertial human body motion capture system of MEMS are designed from the attitude measurement unit, the sensor error is corrected.Attitude calculation and attitude compensation fusion are deeply studied.In this paper, the whole frame of the human motion capture system is designed firstly, and the attitude measurement unit is designed to obtain the acceleration, angular velocity and magnetic field components according to the requirements of the system.Considering the scalability and real-time of the system, a communication protocol supporting asynchronous acquisition and synchronous aggregation of multiple measurement nodes is designed based on the mechanism of dual cache and time division multiplexing.At the same time, after analyzing the random drift in gyroscope, aiming at the difficulty of directly quantifying the gyroscope, the ARMA model is constructed, and the drift of gyroscope is effectively eliminated by Kalman filter.Aiming at the motion attitude analysis of human motion capture system, this paper presents a multi-sensor compensation fusion algorithm based on the experimental comparison of gyroscope integral and magnetic sensor tilt compensation.The attitude fusion is realized based on gradient descent method, complementary filtering and Pi adjustment, and the fusion accuracy is tested in different states.The test results show that the two fusion algorithms achieve the static accuracy of 0.5 掳and the dynamic accuracy of 2 掳.Compared with another common fusion method, Kalman filter, these two algorithms can achieve the corresponding accuracy.And under the same precision, the complexity is smaller and more suitable for low power embedded systems.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212.9

【参考文献】