Visual target tracking method of full-convolution integral type and regression twin network structure

Abstract

A visual target tracking method of a full convolution class and regression twin network structure comprises the following steps: (1) according to the position of a target in an image, cutting a targettemplate image and a search area image in an original training set, and forming a training data set by cut image pairs; (2) establishing a full convolution twin network to extract image features; (3)establishing a classification regression network; (4) in response to the fact that each pixel point on the image has a corresponding foreground score and a predicted bounding box, calculating the total score of each pixel point by combining the information of the foreground score and the information of the bounding box, wherein the pixel point with the highest total score is the center of the tracking target; and (5) training the full convolution twin network and the classification regression network by using the training data set to obtain the trained full convolution twin network and the classification regression network, calculating a score graph of a target in the to-be-tested image sequence by using the trained networks, and performing target positioning based on the score graph. According to the invention, the tracking precision and speed are improved.

Description

technical field [0001] The method relates to the field of visual object tracking, and more specifically, relates to a visual object tracking method with full convolution classification and regression Siamese network structure. Background technique [0002] Visual object tracking is a basic research problem in the field of machine vision, which is widely used in intelligent monitoring, human-computer interaction and unmanned driving and other fields. Although the research of object tracking has made great progress, however, object tracking is still a very challenging task in practical applications. Because in practical applications, the tracked target will inevitably encounter illumination changes, scale changes, background interference, target occlusion and target deformation. [0003] Traditional object tracking methods can be divided into two types: generative-based tracking and discriminative-based tracking. The generative tracking algorithm constructs a model that can ...

AI Technical Summary

Problems solved by technology

[0008] The tracking method based on the region proposal network uses anchor frames for region proposals. Although this approach can effectively use deep feature information and avoid time-consuming repeated calculations, the setting of parameters such as the number, size, and aspect ratio of anchor f

Images