A satellite video target tracking method based on high-resolution Siamese network

Abstract

The invention discloses a satellite video target tracking method based on a high-resolution twin network. The invention adopts a lightweight parallel network to obtain a high-resolution representation of a small target, and realizes robust and real-time tracking. And a pixel-level fine-tuning model is proposed based on tracking. It uses inter-frame motion information to establish a Gaussian mixture model, detects changed pixels, and obtains a binary mask image. The MeanShift algorithm is used to iterate repeatedly on the mask image to achieve fine adjustment at the pixel level. At the same time, taking into account the complementarity of surface features and motion features, the tracking position and fine-tuning position are further adaptively fused, and finally more accurate tracking is obtained.

Description

technical field

[0001] The invention belongs to the technical field of satellite video target tracking, and in particular relates to a satellite video target tracking method based on a high-resolution twin network. Background technique

[0002] Object tracking is an important branch of computer vision science. It has a wide range of applications in video surveillance, intelligent transportation, human-computer interaction, military field, and robot visual navigation. Its purpose is to realize target tracking and positioning through a certain similarity measure and matching search method. So far, the development of target tracking technology in traditional video sequences is relatively perfect. For different application scenarios and requirements, researchers have designed and developed a variety of target tracking methods. Recently, deep learning-based Siamese networks have attracted much attention in traditional tracking due to their amazing accuracy and speed. These Sia...

Images