Occlusion-aware Visual Object Tracker using Kernelized Correlation Filter

The core of many modern trackers is the recognition classifier, whose task is to distinguish between the target and the target's surrounding environment; For the natural change of the image, the classifier uses the result of transform scaling for training; Such sample sets are rife with redundancy and any overlapping pixels are restricted to being identical. This method performs admirably in terms of both tracking effect and tracking speed. To collect samples, the cyclic matrix is used, and the fast Fourier transform is performed to speed up the method. In this algorithm, we first use KCF (Kernel Correlation Filtering) as the basic framework to judge the tracking accuracy of each frame. If tracking confidence is found to decrease, the target is considered to be blocked. To this end, we used Kalman Filter to make educated guesses about the next move of the system for auxiliary tracking, and we tested it on OTB data sets.