A Visual Object Tracking Method Based on Object Analysis

Abstract

The invention discloses a visual object tracking method based on object analysis. The method comprises steps: by learning the color histogram model of foreground and background region online, calculating a region targeting map of the current frame image, locating the location of multiple tracking anchor points, and determining multiple search areas according to the size of the target, using the correlation filter tracking algorithm to obtain the response map of each search area, according to the entropy regularization loss function and the tracking response map of multiple search areas, selecting the best search area, and finally determining the location of the target. As the invention does not carry specific target type information, compared with other existing target tracking algorithms,the method has good performance in solving a moving target platform problem and a background occlusion problem, and has important guiding significance for the research and practice of visual object tracking algorithm.

Description

technical field [0001] The invention relates to the cross technical field of digital image processing and machine learning, in particular to a visual target tracking method based on target analysis. Background technique [0002] Visual object tracking has very important applications in current human-computer interaction, robotic systems, aerospace and urban video surveillance. The commonly used target tracking methods are as follows: [0003] (1) Target tracking method based on image global detection [0004] The target tracking method based on image global detection is mainly divided into two types, that is, the target tracking method based on key point matching and the tracking algorithm based on targetness. The target tracking method based on key point matching usually extracts key points with invariant features at the target position, expresses the target template as a collection of some key points, extracts key points in the current frame full image, and compares them...

AI Technical Summary

Problems solved by technology

[0007] The target tracking method based on global image detection is difficult to real-time due to the large search area and high computational complexity.

In addition, background objects in the global image may also interfere with the tracker, which may easily lead to tracking failure

Although the target tracking method based on local image detection can well solve the problems of background interference and high computational complexity, the tracking result of the algorithm is seriously affected by the predicted position of the target in the previous frame. If the target position prediction result of the previous frame appears deviation, it is very likely to cause the tracking results of all subsequent frames to fail, which limits the practical application range of such methods

[0008] To sum up, although there are already many related methods for visual object tracking, it is difficult to apply them to real and complex natural scenes due to the accuracy, robustness and real-time performance of the algorithm.

Images