基于扩展Kalman预测模型的双目立体视觉图像特征点跟踪方法

Tracking method for stereo vision image feature points based on extend Kalman prediction model

针对双目立体视觉动态测量中图像特征点快速跟踪问题,提出了一种利用Kalman预测模型进行图像点跟踪的方法。该方法以图像特征点坐标值为观测向量,将坐标值和图像特征点运动速度作为状态向量,首先建立了线性Kalman预测递推模型,在此基础上引入了极线约束条件,进一步建立了图像特征点非线性扩展Kalman预测递推模型。实验表明该方法得到的特征点的预测轨迹和真实轨迹符合度好,预测精度高、速度快、通用性强。

Aiming at the problem of rapid tracking for stereo-camera image feature points, a forecast method base on extend Kalman prediction model is proposed. It defines that the observed vector is the coordinates of the image feature points,and the state vector is the coordinates and its speed. Therefore a linear Kalman prediction iterative model is derived. Based on that,a nonlinear extend Kalman prediction model is proposed when taking into account the epipolar constrains in stereo camera vision system. At last experimental results prove that it＇s well accordant between the prediction trace and the real one and this method has the advantages of high forecast precision, rapid tracking speed and wide availability.