Visual target tracking method and system

Abstract

The present invention relates to a visual target tracking method and system. The visual target tracking method includes: acquiring a plurality of historical target templates and historical position information of the current video sequence of the target to be tracked; according to each of the historical target templates, from the Determine multiple sets of target template images and search areas in the current video sequence of the target to be tracked; predict the predicted position information of the target template images in the search area according to the target positioning model, and each set of target template images and search areas; The network model, according to the predicted position information of the target template image and the historical position information, determines the target position forecast revenue value of the target template image; compares the target position forecast revenue values ​​of each target template image, and determines the target position forecast revenue The predicted position information of the target template image with the largest value can accurately determine the predicted position information of the current frame image of the target to be tracked.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a visual target tracking method and system. Background technique [0002] Visual object tracking is an important research direction in computer vision. In each frame of a video sequence, the tracking algorithm needs to determine the position and scale information of the target to be tracked. Since the apparent characteristics of the object to be tracked are only given by the first frame, and various factors such as illumination changes, scale changes, occlusion, and motion blur will occur during the tracking process, the tracking algorithm not only needs to have a strong sense of environmental changes Robustness, and it is necessary to model the apparent characteristics of the target to be tracked with strong discriminative power. Under such conditions, traditional hand-crafted feature-based object tracking algorithms perform poorly. [0003] In recent years, deep lea...

AI Technical Summary

Problems solved by technology

[0004] The existing visual target tracking algorithms based on deep learning or reinforcement learning still have the following deficiencies: First, the target tracking algorithm based on reinforcement learning can only regress the tracking results through multiple discrete actions, not only the tracking efficiency is low, but also the discrete actions lead to The tracking accuracy is poor; secondly, the existing tracking algorithms based on deep learning only use discrete two-frame images for training in the training phase, instead of training on continuous video sequences, which will lead to a large cumulative error of the tracking algorithm , can not track for a long time; finally, these tracking algorithms do not have an effective target template update strategy, as the tracking time becomes longer, the cumulative error gradually increases, eventually leading to tracking failure

Images