Multi-target tracking method based on color and distance clustering

Abstract

The invention provides a multi-target tracking method based on color and distance clustering. According to the method, the influence caused by tracking target deformation can be reduced, the degree of distinction between the targets can be increased and the degree of dependence on the detector can be reduced. According to the method, position information of successive frames and histogram information of the targets are combined, foreground points are clustered and the score of the foreground points is calculated and finally the target position is located, and the histogram characteristics and the position information of the targets are continuously updated.

Description

technical field [0001] The invention relates to the field of digital image processing, more specifically, to a multi-target tracking method based on color and distance clustering. Background technique [0002] Video multi-target tracking technology, that is, for multiple moving targets given in a video sequence, find out their corresponding positions and motion trajectories in each frame, and continue until the end of the video or the target leaves the field of view. Multi-target tracking is of great value in the field of artificial intelligence, such as detecting the dynamics of multiple targets in video surveillance in public places, and in the military field for missile defense, ocean surveillance, and battlefield surveillance. In business, it is used for passenger flow statistics. However, due to the correlation between multiple targets and the complexity of the background, the realization of multi-target tracking is quite complicated. At present, multi-target tracking...

AI Technical Summary

Problems solved by technology

[0005] However, there are still many deficiencies in the existing multi-target tracking technology

The biggest limitation of the method based on detection and matching is that the quality of the tracker is largely affected by the quality of the detector. If the detection effect of the detector is not good, it will directly lead to poor tracking effect, and the efficiency of many detectors Low, unable to meet real-time needs

Moreover, detection-based tracking generally only detects a certain type of object, so the tracker can only track a certain type of object

The online learning method based on the template response, although the speed can be achieved very fast, but a single template cannot meet the movement changes of the target under the camera, such as the deformation of the person under the camera, etc., and the predicted target position is used as a new training set. When updating the template, it is easy to accumulate errors and cause the drift of the template

Images