Target tracking method and training method based on matching-regression network

Abstract

The invention discloses a target tracking method and training method based on a matching-regression network, which are applied to the technical field of image processing, and comprises the following steps: inputting a target to-be-tracked sequence comprising a plurality of frames, and performing target tracking on each frame of image in a target search area; outputting a target center position according to the center matching network, obtaining a feature map output by the last convolution layer of the center matching network according to the determined target center position, and taking the feature map as the input of a boundary regression network; and by the boundary regression network, performing central point divergence according to the input feature map and the target center, determining the boundary position of the target center, and outputting the height and width of an anchor box. According to the method, the center position of the target is determined by using the twin network,and a more accurate anchor box with a variable aspect ratio is output by using two layers of LSTM networks in combination with the time sequence feature information of historical frames in the boundary regression network, so that the accuracy and robustness of target tracking in a video sequence are improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a matching-regression network-based target tracking method and training method. Background technique [0002] Object tracking is an important problem in the computer field. It is widely used in tasks such as automatic driving, video annotation, and pose estimation, which greatly saves computing resources. Compared with other computer problems, such as face recognition, target detection, instance segmentation, etc., the difficulty of target tracking is that the prior knowledge of the tracking target is less (only the appearance image features of the first frame of the object), and it is impossible to pass some offline It is a challenging task to effectively enhance the adaptability to arbitrary objects by means of methods. [0003] After Convolutional Neural Networks (CNN) is applied to the target tracking task, it shows its powerful ability. The convolutional units of ...

AI Technical Summary

Problems solved by technology

At present, the existing problems of the tracking algorithm based on the Siamese network are: (1) The prior knowledge of the tracking target is less, and only the appearance image features of the object in the first frame; (2) In the tracking algorithm based on the Siamese network, the video sequence The first frame of the first frame is used as a template, and only compares the characteristics of the template and the specific area of ​​the current detection frame, without using the timing information between frames in the video sequence; (3) After obtaining the center position of the target, the output of the anchor box is just using Determined by several fixed scales, the adaptability to target deformation is poor

Images