A Video Target Tracking Method Based on Compressed Irregular Block LBP

Abstract

A video target tracking method based on compressed irregular block LBP. This method uses the compressed and sampled irregular block LBP feature vector to describe the tracked target or candidate target, and uses the particle filter framework to search for candidate targets. The Naive Bayesian classifier discriminates whether the compressed feature vector of the candidate target is the target tracking result. This method not only promotes the processing speed of video object tracking, but also maintains accurate tracking effects in various complex scenes such as heavy illumination changes and pose changes, perspective rotation and sudden movements, background clutter, and similar object interference.

Description

Technical field [0001] The invention relates to the field of moving target tracking in a video frame sequence, in particular to a video target tracking method based on compressed irregular block LBP. Background technique [0002] Video target tracking refers to analyzing the motion parameters and trajectories (such as position, size, shape, speed, acceleration, etc.) of a specific target from the video sequence with the help of target characteristics (such as color, texture, shape, etc.), which is the core of the computer vision system One of the tasks is that it has broad application prospects in many fields such as intelligent video surveillance, human-computer interaction, medical diagnosis, and robot navigation. However, various complex scene factors such as illumination changes, shadows, occlusions, sudden motions, and background chaos have brought great challenges to video target tracking technology, and video target tracking methods that are both accurate and fast have att...

AI Technical Summary

Problems solved by technology

Therefore, the disadvantages of the MB-LBP feature mainly include the following two: ①A MB-LBP image area cannot involve all possible scale image areas; ②Compute all possible MB-LBP features of a tracking target area to form a high-dimensional MB-LBP feature Vectors are time consuming

[0035] (2) Inadequacies of compressing Haar-like feature vectors

However, many literatures have confirmed that the performance of ③Haar-like features is far inferior to that of MB-LBP features in video target tracking applications in scenarios such as illumination changes, scale changes, and target rotations.

The FCT method assumes that the Haar-like feature vector dimension n=10 6 ~10 10 , ④ cannot accurately express the dimension of the feature vector of the tracked target area

The sparsity of the measurement matrix R is too high, that is, the number of non-zero elements on each row vector is too small, resulting in the loss of the information of ⑤ compressing the Haar-like feature vector

[0037] (3) Inadequacies of coarse-to-fine search strategy

Compared with the commonly used particle filter search strategy, ⑥The candidate targets generated by the coarse-to-fine search strategy in a large number of impossible directions will cause unnecessary computational overhead

Images