Target tracking method and device

Abstract

The invention discloses a target tracking method and device. The method includes the first step that in an initialized frame, a special mapping link is designed to convert an image into multi-vector mixing features; the second step that extracted multi-vector mixing features are adopted to train a multi-vector correlated filter; the third step that aiming at subsequent sequence images, the trained multi-vector correlated filter is adopted to generate confidence maps of all search locations; the fourth step that the peak value of the confidence maps is selected as a target position, multi-vector mixing features are extracted from the target position through the method the same as the method in the first step, and on-line updating is performed on the correlated filter. By the adoption of the target tracking method and device, a target without prior information can be tracked, high robustness can be achieved on illumination, rotation, shielding and other challenging factors, and quick tracking can be achieved.

Description

technical field [0001] The invention relates to computer vision technology, in particular to a target tracking method and device. Background technique [0002] Object tracking has been studied for many years, and many excellent algorithms have been proposed to handle a variety of different tasks. Object tracking is still a very challenging research task in the field of computer vision because objects are affected by various factors such as illumination, deformation, and occlusion. It is widely used in real life, including video surveillance, human-computer interaction, and video retrieval. [0003] Single object tracking is the most common tracking problem, where most algorithms track an unspecified object without any prior information available. The initial state of the target (position information and scale information) is given in the first frame of the video, and the purpose of tracking is to evaluate the state of the target in subsequent video frames and calculate the...

AI Technical Summary

Problems solved by technology

The method of online update greatly affects the ability of continuous tracking of the target. If an appearance model is updated with obvious noise, errors will accumulate, and eventually lead to obvious drift or even tracking failure.

[0012] To sum up, in the existing target tracking methods, most of the methods sacrifice the speed in order to improve the tracking accuracy, and cannot meet the real-time requirements of 30Hz.

Even though some methods are fast in calculation, the tracking accuracy is low due to the simple design

Images