Target tracking method and device

The invention discloses a target tracking method and device. The method comprises: selecting one frame of image in a to-be-tracked image as a template frame, wherein the template frame comprises a tracking target; extracting the first feature of the tracking target in the template frame and the second feature of other images in the to-be-tracked image through a twin neural network based on a feature pyramid network; identifying the background picture in the to-be-tracked image through classification branch of a region recommendation network; and determining the accurate position of the tracking target in the to-be-tracked image through the regression branch in the region recommendation network. According to the invention, the twin network is used as a feature extraction network to acquirethe feature map fusing the bottom-layer features and the high-layer features, so that the accuracy of the bottom-layer features and the robustness of the high-layer features of the acquired feature map are guaranteed; and the