Method for motion tracking based on predication

Abstract

The invention discloses a method for motion tracking based on predication. The method comprises the following steps that movable objects are initialized and motion tracking is conducted on a motor point cluster layer; the movable objects are tracked, the position of each movable object at the current moment is predicated according to the position of the movable object at the last moment in a predication stage, and which movable objects current frame motor point clusters belong are found in a correction stage; the movable objects disappear and the target change trend is corrected. Due to the fact that the motor point clusters are tracked through the prediction method and the correction method, the problem that one movable object is divided into multiple communication zones when the prediction position of the movable object is searched for the motor point clusters is solved. Due to the fact that motion tracking based on predication is adopted in the method, the defects that the tracking mode of a Kalman filtering system is large in calculated amount and can not be well adapted to actual motion change are overcome and a clear result of complex movable object tracking can be obtained.

Description

technical field [0001] The invention relates to a motion tracking technology, in particular to a motion tracking method based on prediction. Background technique [0002] Object detection technology is a hot spot in current research. After more than 20 years of development, target detection technology has made great progress and has become an important topic in the fields of pattern recognition, image processing, machine vision, and guidance. The ultimate goal of target detection is to accurately separate the target from the background in the video image sequence, then extract the background on the grayscale image sequence, and then extract moving point clusters, motion tracking, etc. [0003] In image processing for motion tracking, a connected region is called a motion point blob. In a moving point-cluster binary image, there are many connected areas, no matter whether the moving object is split into multiple connected areas or the moving object is only a part of a conne...

AI Technical Summary

Problems solved by technology

[0005] However, there are many problems in using Kalman filter for motion tracking: First, the covariance matrix of the system noise vector is necessary, and the system noise is caused by the acceleration of moving objects, and this covariance matrix is ​​difficult to determine

The covariance matrix of the observation noise is also necessary, but it is difficult to quantify

Secondly, the observation results are obtained by two-dimensional search method, which is a very time-consuming method because the number of image pixels is very large

Finally, when calculating the Kalman gain, it is necessary to calculate the inverse of the matrix. If the inverse of the matrix does not exist, it is necessary to calculate its generalized inverse, which is complex and time-consuming

Moreover, there are multiple moving objects in each frame of the image, and each moving object needs to calculate the inverse of the matrix once, and the calculation overhead is very large.

Images