一种用于人体运动捕获的自适应混合滤波融合算法

摘要

针对基于惯性传感器的人体运动捕获系统存在陀螺漂移和噪声干扰等问题,提出一种多元传感器信息融合的自适应混合滤波融合算法.算法首先利用快速高斯牛顿法对加速度计和磁力计数据进行姿态信息迭代估算,用四元数将参考坐标系中的加速度和磁场强度分量转换到载体坐标中,将转换后的值与当前时刻测量值的差值代入高斯牛顿迭代算法中用于四元数的实时值估计,通过确定搜索步长的最优值来缩短迭代次数,提高算法收敛速度.设计自适应的互补滤波器将高斯牛顿法解算的姿态信息作为观测矢量对陀螺漂移进行补偿,分别使用高通滤波器和低通滤波器处理陀螺仪数据和高斯牛顿算法优化过后的加速度计、磁力计数据.在互补滤波器中引入重力矢量及地磁参考矢量自适应调节滤波器参数用于实时调整不同算法的权重大小,融合后输出最终的姿态信息,实现最优估计.进行实验对比分析本算法和其他算法融合效果,结果表明,本算法有效降低陀螺累积误差、线性加速度及磁场对解算精度的干扰,磁干扰状态下误差为0.94°,自由运动状态下误差为1°.对比扩展卡尔曼滤波融合算法,本文算法执行时间缩短25%,有效提升了运动捕获系统的性能.%In order to solve the problem of gyro drift and noise disturbance in human motion capture system,an adaptive Gauss-Newtor/complementary filter fusion algorithm based on the information fusion of multi sensors was proposed.The Gauss-Newton method was first adopted to process the data measured by accelerometer and magnet.Then components extracted from the acceleration and magnetic intensity were converted from reference coordinate system to carrier coordinate by quatemion,and the difference between converted value and current time value was introduced into the Gaussian Newton iterative algorithm for real time value estimation of the quaternion.The time of iteration was shortened and the convergence speed of algorithm was improved afterwards by optimizing the optimal value of the step length.Furthermore,an adaptive complementary filter was proposed to compensate the gyro drift and attitude information solved by Gauss-Newton method was used as observation vector.Then,the high-pass filter and the low-pass filter were used to process the gyroscope data,accelerometer and magnetometer data optimized by Gauss-Newton algorithm.To obtain the optimal estimation after fusion,the gravity vector and geomagnetic reference vector adaptive filter parameters were used to adjust the weights of different algorithms in real time.The results showed that the algorithm effectively reduced the interferences caused by gyro divergence,linear acceleration and magnetic field.Under the condition of magnetic interference,the value of error was 0.94°.Under the condition of free movement,the value of error was 1 o.Compared with extended Kalman filter fusion algorithm,the algorithm execution time is reduced by 25%.Therefore,the performance of motion capture system was effectively improved.