Accurate target tracking method on condition of severe shielding

Abstract

The invention provides a Gaussian sparse representation cooperation model which is used for target tracking on the condition of severe shielding. The accurate target tracking method is characterized in that a manner of combining sparse coding and LLC coding is utilized for performing sparse representation on candidate samples. According to method of the invention, a sparse solution can be easily acquired and furthermore a high-precision reconstruction error is obtained; furthermore prior probability is added into the model so that the sample at the periphery of the next target frame can be more easily used as a final tracking result. Through a large number of experiments for comparing with other methods, the accurate target tracking method can realize better target tracking effect on the condition of severe shielding.

Description

technical field [0001] The invention belongs to the field of pattern recognition and machine learning, and can be used for object tracking, especially for the situation that the tracking object is seriously blocked. Background technique [0002] There are many ways to classify target tracking methods. In chronological order, they can be divided into tracking algorithms that appeared in the last century and the beginning of this century, such as Cam Shift, Mean Shift, Kalman Filtering, Optical Flow, and Particle Filtering. Many non-deep learning tracking algorithms, such as KCF, SCM, TLD, Struck, etc. Until 2012, Hinton, one of the founders of deep learning, promoted the development of deep learning. In recent years, many deep learning target tracking algorithms have emerged, such as MDnet and CNT. Different types of target tracking methods expose different problems. [0003] 1. Among them, the tracking method that needs to be pre-trained in advance is prone to the deviatio...

AI Technical Summary

Problems solved by technology

[0003] 1. Among them, the tracking method that needs to be pre-trained in advance is prone to the deviation of the predicted target position, and this type of method has a dependence on the training data set; while the offline tracking method can globally optimize the tracking path, and can be used on the test sequence to Scanning forward or backward, but there are relatively few applications, so the method of online tracking is the main research direction

[0004] 2. In actual situations, the combination of generative model and discriminative model performs better, mainly because the generative model tracking method is prone to errors when the background is more complex and there are many interference factors, and it is difficult to correctly judge the target direction of movement

When the target is heavily occluded, the original features of the target partially disappear, and these deep features cannot be matched, so the tracker can track the target in a wide range search, which can easily lead to the loss of the target or have a large center point error

[0006] 4. Figure 8 It is the most common and difficult occlusion problem in the field of target tracking research. When the tracking target is occluded, the original features of the target become incomplete. At present, it is difficult for many trackers to capture the target. When the target is lost, the tracking algorithm is difficult to find. back to the target, resulting in tracking failure

Images