Abstract: Aiming at the problem that it was difficult to achieve high-precision and high frame rate tracking under limited computing resources, we proposed a lightweight  siamese network target  tracking algorithm based on ananchor free.   Firstly, the modified lightweight network MobileNetV3 was used as the backbone network to extract features, and reduced parameters and computation of the network  while maintaining deep feature expression capability. Secondly, for traditional cross-correlation operation, we proposed deep cross-correlation module for graph cascading optimization, which highlighted important information of target features through rich feature response graphs. Finally, feature sharing was used  to reduce parameters and computation to improve tracking speed in the anchor classification regression prediction network. Comparative experiments were conducted on two mainstream datasets OTB2015 and VOT2018, the experimental results show that the algorithm has a significant accuracy  advantages compared to  SiamFC tracker, and is more robust in complex tracking scenes. At the same time, the tracking frame rate can reach 175 frames/s.

Key words: target tracking, siamese network,  , lightweight network MobileNetV3, cross-correlation module, anchor free