CN104680521A

CN104680521A - Improved background modeling and foreground detecting method

Info

Publication number: CN104680521A
Application number: CN201510065105.XA
Authority: CN
Inventors: 樊滨温; 刘晓炯; 王明江; 曲中鑫; 卢婷舒; 刘明
Original assignee: Shenzhen Graduate School Harbin Institute of Technology
Current assignee: Shenzhen Graduate School Harbin Institute of Technology
Priority date: 2015-02-06
Filing date: 2015-02-06
Publication date: 2015-06-03
Anticipated expiration: 2035-02-06
Also published as: CN104680521B

Abstract

The invention provides an improved background modeling and foreground detecting method. The method comprises the steps of establishing a background model at first, then judging that pixel points of a current frame belong to background points or foreground points, updating a background set, judging each image frame pixel by pixel through the former process, finally obtaining a binary image where foreground is separated from background and accordingly dividing foreground moving targets from monitoring video. The method overcomes the defects in a traditional target detecting method and improves detecting adaptability, stability and timeliness.

Description

A kind of background modeling of improvement and foreground detection method

Technical field

The present invention relates to image intelligent detection, recognition technology field, particularly relate to a kind of method of background modeling and foreground target detection.

Background technology

In intelligent monitoring, moving object detection is significant in the daily productive life of people, its objective is and is split from background by moving foreground object.But in a practical situation, monitored environment is many uncertain factors abrim, background is often easy to be subject to various impact, and the such as impact such as illumination variation, dynamic background, shade, partial occlusion, these unfavorable factors bring very large challenge to anaphase movement Target detection and identification.Therefore, to various complex situations analysis modeling, reduce the adverse effect that extraneous complex environment factor causes target detection recognition result as far as possible, reduce erroneous judgement, the adaptability, robustness etc. that improve algorithm just become very important.

In foreground detection field, main flow algorithm mainly contains frame difference method, optical flow method and background subtraction method etc.Frame difference method has good real-time, and its background is not assembled, and renewal speed is fast, algorithm is easy, calculated amount is little.The deficiency of algorithm is neighbourhood noise comparatively responsive, and the selection of threshold value is quite crucial, selects the too low noise be not enough in oppressive image, too high, has neglected change useful in image.Likely occur abstract in target internal, action target cannot be extracted all sidedly.Optical flow method does not need to obtain scene information in advance, just can target be detected, but noise, multiple light courcess, shade and the factor such as to block and can impact the calculating convergence of optical flow field; And optical flow method calculation of complex, be difficult to the object reaching process in real time.Background subtraction method must be had powerful connections image, present image is utilized to deduct reference background image, and background image must be change along with illumination or external environment condition and real-time update, therefore the key of background subtraction method is background modeling and renewal thereof, and the quality of segmentation result depends on selected background modeling method to a great extent.

Due to the property complicated and changeable of actual environment, many traditional background modeling methods all have its limitation, and therefore many method for building up being proved to be the background model of robust are suggested.These methods not only improve Detection results to a great extent, and improve adaptability, the stability and computing velocity etc. of algorithm.

Summary of the invention

In order to solve the problems of the prior art, the invention provides a kind of background modeling and foreground detection method of improvement, from monitor video, splitting foreground moving object.The method overcome some shortcomings of conventional target detection method, improve and detect adaptability, stability and real-time etc.

The present invention is achieved through the following technical solutions:

The background modeling of improvement and a foreground detection method, is characterized in that: said method comprising the steps of:

Steps A 1: choose front N two field picture, calculate the average I of each pixel position ⁿ(x, y), for each pixel sets up respective background collection, the background element of set of each pixel element is randomly drawed form by the past N two field picture with the pixel value of this pixel same position, and each background is concentrated and comprised M element.

Steps A 2: judge that each pixel of present frame belongs to background dot or foreground point, if the number that the similarity of the element that current pixel background is concentrated and this pixel is less than a setting threshold value th1 is greater than T1, then this point is judged as background dot, otherwise is judged to be foreground point.

Steps A 3: if certain pixel is detected as foreground point, do not upgrade background collection, if be detected as background dot, upgrades the background collection of this position pixel.

Further, described steps A 3 specifically comprises the following steps:

Step B1: if certain point is judged to background dot, first judge the speed that this pixel changes, in the continuous K two field picture before namely finding out present frame, this position pixel is judged to the pixel of background dot, K<N, and adds up these background dots and average I ⁿthe difference of (x, y), if the number that absolute difference is greater than the point of setting threshold value th2 is individual more than T2, then can judge that this current background dot belongs to the fast background dot of change, otherwise current background point belongs to the slow background dot of change.

Step B2: if this pixel being detected as background dot belongs to the slower point of change of background, replace with this pixel value some elements that its background concentrates random selecting; If belong to change of background point faster, then replace with this pixel value m the element that its background concentrates random selecting.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of background modeling of the present invention and foreground detection method.

Embodiment

In order to make object of the present invention, technical scheme and advantage clearly understand, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.

Accompanying drawing 1 is the process flow diagram of background modeling of the present invention and foreground detection method.

First background model is set up: choose front N two field picture, for each pixel sets up respective background collection, the background element of set element of each pixel is randomly drawed form by the past N two field picture with the pixel value of this pixel same position, and each background is concentrated and comprised M element.

Then judge that each pixel of present frame belongs to background dot or foreground point: the similarity (Euclidean distance namely between pixel first judging the element that current pixel background is concentrated and this pixel, as two point (r1, g1, and (r2 b1), g2, b2) between Euclidean distance be if the number that this pixel and its background concentrate the similarity of each element to be less than a setting threshold value th1 is greater than T1, then this point is judged as background dot, otherwise is judged to be foreground point.

Then upgrade background collection: if current pixel is judged as background dot, then continuing to detect this background dot, to belong to change fast or change slow point, if belong to the slower point of change of background, replaces with this pixel value some elements that its background concentrates random selecting; If belong to change of background point faster, then replace with this pixel value m the element that its background concentrates random selecting.

Utilize the background modeling of the present invention shown in Fig. 1 and foreground detection method flow process to carry out judging by pixel to each picture frame, finally obtain the bianry image of prospect and background separation.

Above content is in conjunction with concrete preferred implementation further description made for the present invention, can not assert that specific embodiment of the invention is confined to these explanations.For general technical staff of the technical field of the invention, without departing from the inventive concept of the premise, some simple deduction or replace can also be made, all should be considered as belonging to protection scope of the present invention.

Claims

1. the background modeling improved and a foreground detection method, is characterized in that: said method comprising the steps of:

Steps A 1: choose front N two field picture, calculate the average I of each pixel position ⁿ(x, y), for each pixel sets up respective background collection, the background element of set of each pixel element is randomly drawed form by the past N two field picture with the pixel value of this pixel same position, and each background is concentrated and comprised M element;

Steps A 2: judge that each pixel of present frame belongs to background dot or foreground point, if the number that the similarity of the element that current pixel background is concentrated and this pixel is less than a setting threshold value th1 is greater than T1, then this point is judged as background dot, otherwise is judged to be foreground point;

2. method according to claim 1, is characterized in that: described steps A 3 specifically comprises the following steps:

Step B1: if certain point is judged to background dot, first judge the speed that this pixel changes, in the continuous K two field picture before namely finding out present frame, this position pixel is judged to the pixel of background dot, K<N, and adds up these background dots and average I ⁿthe difference of (x, y), if the number that absolute difference is greater than the point of setting threshold value th2 is individual more than T2, then can judge that this current background dot belongs to the fast background dot of change, otherwise current background point belongs to the slow background dot of change;

3. method according to claim 2, is characterized in that: the value of described m is 3 ~ 5, or the value of described m suitably adjusts according to concrete background environment.