Visual object tracking method based on multiple model integration and structured depth characteristics

Abstract

The invention belongs to the technical field of pattern recognition and computer vision, and discloses a visual object tracking method based on multiple model integration and structured depth characteristics. The method comprises the steps that modeling is conducted on the appearance of a tracking object by adopting multilayer structured depth network characteristics, network extraction characteristics have higher robustness on interference factors such as motion blur; low-level characteristics in structure characteristics can not only distinguish strong interfering targets easily but also make position estimation more accurate; high-level characteristics in the structured characteristics can separate the tracking object from a background easily. Accordingly, visual object tracking is conducted by means of long time and short time model merging, and the precision of position estimation is improved easily through a short-time model; the strong interfering targets which are similar are inhibited and tracked easily through the short-time model. The method has the advantages of being high in precision and robustness; the method can be used for application such as video monitoring, roadtraffic condition analysis and human-computer interaction.

Description

technical field [0001] The invention belongs to the technical field of pattern recognition and computer vision, and in particular relates to a visual target tracking method based on multi-model fusion and structured depth features. Background technique [0002] Visual object tracking is one of the most fundamental problems in the field of computer vision. With the development of information technology, video has become an important information carrier. A large amount of video data leads to the demand for automatic video analysis and processing, and object tracking is one of the key technologies that need to be solved. Since the input of visual object tracking is only images in video, it is easily affected by the quality of video capture. According to the video classification of several publicly available test databases, the main factors affecting video quality and tracking difficulty are: illumination changes, scale changes, occlusion, motion blur, complex backgrounds, rot...

AI Technical Summary

Problems solved by technology

Among the above two methods, the method based on the generative model does not use the information of the background, so it is difficult to separate the target in the presence of a complex background, and there is a high probability that it will track an unrelated object or background; while the method based on discrimination Although the linear model method has a strong ability to separate the target and the background, it is limited by the small number of online samples and is prone to overfitting.

[0003] To sum up, the problems existing in the existing technology are: the appearance model based on a single feature is difficult to fully describe the tracking target, cannot accurately distinguish the target from the background, and is likely to lose the target after a period of tracking

In addition, due to the small number of samples available for online update models, the current mainstream tracking methods based on discriminative models are prone to overfitting, resulting in a decrease in the estimation accuracy of the target position.

Images