Anchorless target tracking based on region generation

With the development of artificial intelligence in recent years, target tracking algorithms based on deep learning have been more and more widely used in the fields of unmanned aerial vehicles (UAVs), autonomous driving, sports events, and public safety. Although current target tracking algorithms have multiple solutions for problems such as occlusion, deformation, jitter, and fast movement, the accuracy of the tracker's position prediction is greatly reduced when the target undergoes deformation in a complex environment. In this paper, we propose to introduce an anchorless method based on a regional generative twin network to solve the problem of long computation time consumed by the model while performing the tracking task and the problem of high tracking failure rate when the object undergoes deformation. In the branch of predicting the object position, by directly predicting the distance between the centre point and the perimeter of the border, the number of parameters is five times less compared to the original anchor point-based target tracking algorithm, effectively shortening the computation time required for target tracking, while the accuracy remains almost the same. Experimental results on the VOT dataset show that this method is higher than the traditional method model in terms of accuracy and speed.