Video multi-target long-range tracking method

Abstract

The invention relates to a video multi-target long-range tracking method. The method comprises the following steps of: step 1, target features of a large number of sample pictures to be tracked are extracted, at first, a M- tree is adopted to represent a feature set of word frequency vectors of feature words of a target sample, and then, video retrieval is adopted to perform M- tree retrieval on targets to be matched, and finally, an affine transformation relationship is adopted to verify retrieval results, and therefore, initial targets can be matched; step 2, a short-range tracker of optical flow tracking is adopted to track and detect the initial targets; step 3, a three-dimensional target model of online learning is adopted to update the targets so as to construct a multi-form multi-perspective target model; and step 4, and short-range tracking and target redetection are combined so as to realize long-range tracking. According to the video multi-target long-range tracking method of the invention, the three-dimensional target model of online learning is adopted, and therefore, problems caused by constantly changing target objects in network video tracking can be solved, and problems existing in detection of targets of which image features change constantly can be solved; and a tracking framework of the short-range tracking and target redetection is adopted, such that long-range racking of targets can be realized. The video multi-target long-range tracking method can be widely used in fields such as network information security and multi-target detection and video multi-target tracking.

Description

technical field [0001] The invention relates to a video multi-target long-distance tracking method. Background technique [0002] At present, online video has gradually become an important source of information in people's lives, and its influence has gradually surpassed traditional TV broadcasting, which has gradually begun to develop into online video. On the other hand, the freedom of online video uploading has gradually brought to the fore the issue of content security, which has become the focus of online video and has inhibited the healthy development of online video. Analyzing and managing the content of network video information has become an important task of network information security. [0003] Video target tracking, by analyzing a series of historical noisy observation information in the video sequence, estimates the real parameters of the target, such as position, acceleration, speed, trajectory, etc. These observations may come from tracked targets, environm...

AI Technical Summary

Problems solved by technology

The long-range tracking method based on the combination of tracking and detection relies on the detector trained offline, and the detector will not change its attributes during the tracking period. Although the adaptive discriminative tracker can solve the problem of appearance change, it cannot simultaneously contain the multi-morphology of the target. Appearance information of viewing angles, leading to inability to perform robust long-range tracking for real complex network video tracking

In addition, when applied to multi-target detection, the detection time of existing detection methods based on local image features and sliding window detection methods will increase linearly with the increase in the number of sample targets to be detected.

For a large number of image objects, the detection efficiency is very low

Images