Video single-target tracking method combining multi-feature and double-threshold updating strategy

Abstract

The invention relates to a video single-target tracking method combining multiple features and a double-threshold updating strategy, and the method comprises the steps: extracting the combined feature data of a specific region, so as to determine a corresponding detection response diagram, the combined feature data comprises HOG features, CN features and LBP features, and more abundant target feature information can be obtained; and by utilizing a comparison result of a maximum value in the detection response diagram and an average response value and combining a comparison result of a confidence coefficient of a historical information change trend and an average confidence value, judging whether a shielding object intrudes into the video image or not, thereby selecting whether to update a tracking model coefficient or not according to different judgment results. The target is tracked accurately in real time, the misjudgment probability is greatly reduced, the model performance is improved, and the problem that the tracking performance of a conventional tracking method is reduced or even completely invalid is solved.

Images