Background modeling method in target detection [go]

Recently has been doing the prospect of research, the beginning is mainly to do some engineering applications, in order to solve the engineering problems, a lot of Kung Fu, also read a lot of recent articles at home and abroad.      Always want to do a summary, dragged drag, and eventually wrote this very unsuccessful summary. The algorithm of background modeling or foreground detection mainly includes:

1. Single Gaussian (Gaussian model)

Real-time tracking of the human body

2. Mixed Gaussian models (Mixture of Gaussian model)

An improved adaptive background mixture model for real-time tracking with shadow detection

3. Sliding Gaussian average (Running Gaussian average)---single Gaussian

Real-time tracking of the human body

For Gaussian and mixed Gaussian estimation everyone is familiar with, here is no longer tired (mixed Gaussian in the existing background modeling algorithm should be considered relatively good, many new algorithms or improved algorithms are based on some of its principles of the different variants, but the disadvantage of mixed Gaussian algorithm is relatively large, slow speed, light sensitivity);

4. Codebook (CodeBook)

Real-time Foreground–background segmentation using Codebook model

Real-time Foreground-background segmentation using a modified codebook model

On and Codebook algorithm, once did the experiment, the effect can also, later also has many variants, did not carry on the research, but the algorithm to the illumination is also sensitive);

5. Self-organizing background detection (sobs-self-organization background subtraction)

A self-organizing approach to background subtraction for+visual surveillance

For the self-organizing background modeling algorithm, sobs algorithm, the algorithm has certain robustness to illumination, but the map model is larger than the input picture, the computational amount is large, but it can be solved by parallel processing to solve the speed problem of the algorithm, can be tried;

6. Sample consistent background modeling algorithm (SACON)

A consensus-based method for tracking

A consensus-based method for tracking-modelling background scenario and foreground appearance

Sacon-background subtraction based on a robust consensus method

Sacon algorithm is based on statistical knowledge, code implementation, and done experiments, the effect can also, but no further analysis;

7. Vibe algorithm

Vibe

Vibe-a Universal Background Subtraction

Vibe algorithm is a masterpiece of B, online has a ready algorithm available, but has applied for a patent, used for research or can be, the algorithm is very fast, the calculation is small, and the noise has a certain robustness, the detection effect is good;

8. Color information based background modeling method (color)

A statistical approach for real-time robust background subtraction and shadow detection

Based on color information background modeling method, called color algorithm, this algorithm will divide pixel difference into chromaticity difference and brightness difference, have strong robustness to illumination, have better effect, calculate speed is relatively fast, can meet the requirement of real-time, Do a lot of video sequence detection, the effect is more ideal;

9. Statistical averaging method

10. Median filter method (temporal Median filter)

Automatic congestion detection system for underground platform

Detecting moving objects,ghost,and shadows in video streams

Statistical averaging method and median filter method, for these two algorithms, only the statistical averaging method has been implemented, and tested, the application of the algorithm has great limitations, can only be regarded as a theoretical supplement;

One-W4 method

W4.pdf

W4 algorithm should be used in the first practical application of an algorithm, the people can go to see the relevant information, here no longer elaborate;

12. Intrinsic Background law

A Bayesian Computer Vision system for modeling human interactions

The background of the law has not been achieved, see a lot of literature has been explained, and then the algorithm is based on the Bayesian framework, I have been the Bayesian framework is not cold, theoretically very perfect, the practical application is shit (no offense to the Bayesian fans, if offended, please detour, do not spit);

13. Kernel Density Estimation method

Non-parametric model for background subtraction

Finally, the kernel density estimation algorithm, the algorithm should be a relatively robust algorithm, can solve many of the problem of algorithm parameter setting, no need to set parameters should be a big advantage of the algorithm.

Personal views: Sobs, Color, VIBE, Sacon, PDF, etc. can be deeply understood, especially in recent years block-based or region-based, features-based, The algorithm based on hierarchical classification or level trainer can be studied in depth.

Recommend a review article: Evaluation of Background subtraction techniques for Video surveillance

Recommend another website: http://www.changedetection.net/

Background modeling method in target detection [go]