改进梯度下降法的机载云台姿态解算

Attitude Algorithm of Airborne Pan-tilt Based on Improved Gradient Descent Algorithm

针对机载云台计算能力有限、动态性能较差的问题,提出一种基于低成本三轴无刷云台的改进梯度下降法姿态解算。该算法采用四元数描述系统模型,先进行卡尔曼滤波预处理加速度计的输出数据,降低动态误差,后利用基于梯度下降法姿态解算来减小姿态误差,最后线性融合加速度计、磁力计和陀螺仪解算的姿态四元数,从而得到能够适应复杂环境的最优姿态,计算量较小的同时保证机载相机的实时稳定性。通过两组对比实验表明,改进算法较梯度下降法精确度更高,静态、动态性能分别提升15.25%和27.53%,同时与显式互补滤波进行对比,印证了改进算法在云台系统中的可行性。

Aiming at the problem of limited computing power and poor dynamic performance of airborne Pan-tilt, an improved gradient descent algorithm based on low-cost three-axis brushless Pan-tilt is proposed. The algorithm uses quaternions to describe the system model. Firstly, Kalman filter is used to preprocess the output data of accelerometer to reduce the dynamic error. Then, the attitude algorithm based on gradient descent is used to reduce the attitude error. Finally, the attitude quaternions of accelerometer, magnetometer and gyroscope are fused linearly to obtain the optimal attitude which can adapt to complex environment. The real-time stability of airborne camera is guaranteed while the computation is small. By showing that in two groups of experiments, the improved algorithm with gradient descent algorithm has higher accuracy, static and dynamic performance were increased by 15.25% and 27.53%. At the same time, compared with explicit complementary filtering algorithm, it proves the feasibility of the improved algorithm in the Pan-tilt system.