Kernelized Correlation Filtering algorithm fused with Kalman Filter

The KCF algorithm updates the model by sampling each frame of the image. When the object is obscured by the object, the wrong sample will be introduced in the model update, making the tracking effect worse or even the target lost. Aiming at the above problems, KCF algorithm is fused with Kalman filter. But in fact, the target motion in the image is nonlinear due to the translation and rotation of the camera, while Kalman filter can only deal with the linear motion of the target. Through the transformation of world coordinate system - camera coordinate system - image coordinate system - pixel coordinate system, the two-dimensional coordinates of the target in the pixel coordinate system of KCF algorithm are converted into the three-dimensional coordinates in the world coordinate system by combining the distance from the camera to the target and the camera's attitude angle and offset. The nonlinear motion of the target in the image is converted into the linear motion in the world coordinate system. According to the assumed target motion model, the Kalman filter is used to estimate the optimal three-dimensional coordinates of the target, and then the target position is converted into the target position modifier in the KCF algorithm. The experimental results show that the improved algorithm is more robust than KCF when the target is occluded.