双目视觉基于修正卡尔曼滤波器的运动估计

Two-camera System Object Motion Estimation Based on Modified Kalman Filter

研究了目标物体的远程运动估计。首先,建立了一种双目视觉系统的基于卡尔曼滤波器的目标物体运动估计的运动学模型,并且证明了双目视觉系统同步的各自连续两帧图像中至少三个对应图像点能完全确定刚性物体的运动参数和空间位置;然后,通过对状态向量中的速度分量进行再估计,提出了一种修正卡尔曼滤波器对目标物体远程运动估计的算法,与直接卡尔曼滤波器的远程运动估计相比,提高了估计的精度。将该方法运用到一种实时预测的实验中,其结果证明了该算法的有效性。

The method of object long term motion estimation is discussed.A discrete-time dynamic model of two-camera system based on Kalman filter is presented.It proves that three corresponding points of two successive frames can conform the motion parameters and space location of object.A method of object long term estimation using modified Kalman filter by reestimating the velocity vector of system state vector is presented.Compared with object long term estimation using Kalman filter directly,this method improves the estimation accuracy.A real-time algorithm using the modified Kalman filter is presented.The simulations results show the effectiveness of this algorithm.