No self-introduction
Loading
Profile
Follow
codes (1)
Kalman+camshift Tracking program
3.8
Aiming at the problem that CAMSHIFT tracking is easy to lose the target, Kalman filter is added to predict the next possible position, which reduces the probability of tracking loss. The prototype of cam shift algorithm is mean shift algorithm. The core of the latter is to iterate along the gradient direction of a certain function, and then converge to a certain range according to the number of iterations and error threshold, which is the area to be found. This function is the projection image of the color histogram of the target region in the current key frame, which can be understood as a spatial function distribution - color topographic map. Cam shift algorithm is an upgrade of mean shift algorithm, which can iterate with variable window Cam shift is an iterative tracking algorithm based on density function, which is insensitive to motion, so we can't do the correlation analysis between two frames (we need to do this part ourselves). The reason why this algorithm can be used for tracking is the premise assumption of the algorithm, that is, the motion of the target between the connected frames is small, that is, the motion is approximately continuous, so that the density function is continuous, and the local optimal value can be found along the gradient direction, which is the local optimal value, which is the target region. Cam shift algorithm is easy to lose, because the density function may have multiple peaks, and the gradient search is easy to fall into the local optimal value, but not to the global optimal value. The appearance of multi peak is due to the interference of background. Therefore, the selection of density function, that is, the representation of target features, is a very important part.
onlyMe901829 0 0
1
No more~
Follows0
Fans0