Research on target loss early warning of KCF algorithm based on hypothesis test

The KCF algorithm uses the response values of the template and the test sample to calculate the target position. The location of the maximum response value is the target region. However, the KCF algorithm does not set the target loss warning mechanism. The target will be lost when the tracking process encounters complicated conditions such as scale chance, rapid movement, and severe obstruction. At this point, the tracker will update the background information into the template. The accumulated deviation of the template will cause the tracker cannot be correctly identified and tracked when it encounters the target again. To solve this problem, through the statistics of the maximum response value of each frame of the video sequence, it founds that the overall response of the video sequence has a normal distribution trend, and the maximum response value of the frame will fluctuate abnormally in the frames that are lost or are about to be lost. This paper uses the idea of hypothesis testing in mathematical statistics. This paper uses a set of fixed-dimensional response peak data to perform hypothesis testing on the peak value of the current frame response. If the response peak of the current frame falls within the rejection region, the target is determined to be lost or is about to be lost. The experimental results prove that the proposed method can correctly implement the early warning function when verifying the OTB standard test sequence. It can provide reference for when to load redetection after target is lost.