Robust single target tracking method based on instance feature perception

Abstract

The invention discloses a robust single-target tracking method based on instance feature perception, and the method comprises the following steps: 1, model training: training a network model through employing a server, optimizing network parameters through reducing a network loss function until the network converges, and obtaining the network weight of robust single-target tracking based on instance feature perception; and 2, model inference: tracking a target in a new video sequence by using the network weight obtained in the training stage. According to the robust single-target tracking method based on instance feature perception, features between similar targets are distinguished through learning, the discrimination capability of a tracker is enhanced, and the anti-interference capability and robustness of the tracking process are improved. According to the method, the target can be accurately and stably tracked in numerous difficult actual scenes, and compared with other methods, a better target tracking and capturing effect is achieved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a robust single target tracking method based on instance feature perception. Background technique [0002] Deep learning promotes the rapid development of the field of computer vision, and object tracking has important application value and significance in real life, such as robot vision tracking systems, video surveillance systems, etc. From the perspective of feature extraction, object tracking has mainly gone through three stages: traditional manual feature extraction, deep features, and end-to-end deep features. In the first stage, the tracker usually uses color, histogram, etc. to represent the features of the tracked object, and uses support vector machines, correlation filtering, etc. to achieve feature matching. The features extracted by such methods have limited generalization and thus limited tracking performance. In the second stage, deep learning...

AI Technical Summary

Problems solved by technology

[0005] The present invention aims at the problem that in the existing deep learning-based single-target tracking method, the ability to discriminate similar target interference is insufficient, and it is easy to follow the wrong target during the tracking process, a robust single-target tracking method based on instance feature perception is proposed

Images