Dense spatio-temporal context target tracking method based on adaptive model

Abstract

The invention proposes an adaptive model-based dense spatial-temporal context (STC) target tracking method. The method comprises the following steps of S1, specifying and giving a tracking target of an initial rectangular frame of a first frame of a video by a user, performing initialization to obtain an STC conventional template, then immediately performing snapshot storage once on the conventional template to obtain a first historical snapshot template, and adding the first historical snapshot template into a snapshot template group; S2, when a t frame comes, performing tracking and estimation on the t frame by using the historical snapshot template group and the STC conventional template simultaneously; and S3, extracting a highest confidence degree obtained by estimation in the snapshot template group, substituting the highest confidence degree into a formula defined in the specification to perform judgment, if the adaptability of the historical snapshot template is higher than that of the conventional template, performing snapshot rollback on the conventional template by the historical snapshot template, and finally substituting a frame index value into a formula IndexFrame% phi==0 to judge whether an interval threshold of snapshot acquisition is reached or not to acquire a new snapshot, so as to always keep accurate tracking of a target.

Description

technical field [0001] The invention relates to the field of computers, in particular to a dense spatio-temporal context target tracking method based on an adaptive model. Background technique [0002] Computer vision refers to the use of computers and related imaging equipment as the hardware basis to collect image information, and then use computer methods and other software to process the collected image information to obtain the target or semantic content, so as to realize the simulation of biological vision system . Among them, imaging equipment is used as a substitute for visual organs to collect image information of the surrounding environment, and computer methods are used as a substitute for the human brain to process the collected information to obtain scene information and content of interest. [0003] Among them, target tracking in video is one of the important problems in the field of computer vision. It not only needs to overcome many technical difficulties, b...

AI Technical Summary

Problems solved by technology

[0015] 1) First of all, the method based on the time linear structure model is too simple, ignoring the continuity of the change of the target to be tracked on the time axis, while the method based on the Bayesian average model blindly calculates the probability density function of all forward frames. average;

[0016] 2) Secondly, so far, there is no perfect method that can obtain the most perfect results in the prediction and estimation of each frame, especially in the tracking of complex scenes where multiple interferences coexist;

[0017] 3) Also includes potential sampling error

However, since the two models based on the time smoothing assumption ignore the error introduced by the method estimation, the error information will be learned and accumulated in the method template, which will eventually cause the target tracking to drift or even be lost.

[0019] Because the STC method is a method based on the Bayesian average model, there must be inherent defects of the same type of model method that are prone to model drift

Images