Foreground Target Detection 1 (Summary)

Motion foreground object detection has always been one of the difficulties and hot spots in the field of visual monitoring at home and abroad. Its purpose is to extract the changed area from the background image from the sequence image, effective Detection of moving foreground objects is crucial for post-processing such as object tracking, object classification, and Behavior Understanding. To distinguish foreground objects, a critical issue is to determine a very suitable background, the background is understood from the perspective of pixels. Each pixel is either a foreground or a background. Therefore, we must prevent the background from mistakenly accessing the original foreground object, there are currently several common methods, but they both have advantages and disadvantages.

It can be divided into: Background Modeling, frame difference method, and optical flow method.

1: single Gaussian (single Gaussian Model)

Real-time tracking of the human body

 

2: mixture of Gaussian Model (Gaussian Mixture Model)

An Improved Adaptive Background Mixture Model for real-time tracking with Shadow Detection

The adaptive background difference algorithm based on the Gaussian mixture model is similar to the inter-frame difference method. The Gaussian mixture distribution model is used to characterize the features of each pixel in an image frame. When a new image frame is obtained, the Gaussian mixture distribution model is updated in a timely manner. A subset of the Gaussian mixture model is selected to represent the current background. If a certain pixel of the current image frame matches the background subset of the Gaussian mixture model, it is determined as the background, otherwise it is determined as the foreground point. In general, learning and training are used to calculate whether there is any background for moving objects.

 

3: Running Gaussian average (moving Gaussian mean)

Real-tine tracking of the human body

Gaussian mixture is a good algorithm in the existing background modeling algorithms. Many new algorithms or improved algorithms are deformed based on their principles. However, the disadvantage of Gaussian mixture is that the calculation workload is relatively large, slow speed, sensitive to light.

 

4: codebook)

Real-time foreground-Background Segmentation Using codebook Model

Real-time foreground-Background segmentation using a modified codebook Model

The effect of the code book is acceptable. Later versions are available, but they are sensitive to light;

 

5: Sobs-self-organization Background Subtraction (self-organizing background detection)

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

For the self-organizing Background Modeling Algorithm (sobs), this algorithm is robust to illumination. However, the map model is larger than the input image, and the computing workload is relatively large, however, you can solve the speed problem of the algorithm through parallel processing. You can try it;

 

6: sacon (sample consistency Modeling Algorithm)

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

This method is based on statistical knowledge and has good results;

 

7: vibe Algorithm

Vibe-A Universal Background Subtraction

Author's website: http://www2.ulg.ac.be/telecom/research/vibe/

The vibe algorithm is a different algorithm from the traditional method. The author has applied for a patent and can do research. The background model can be initialized with a frame of image. The method requires a small amount of computing and is fast, it can be embedded into the camera to resist camera jitter and be robust to noise. The detection effect is very good;

 

 

8: color (Background Modeling Method Based on Color information)

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

Color information-based background modeling method, referred to as the color algorithm. This algorithm divides pixel differences into chromaticity differences and brightness differences, which are robust to illumination and have better results, the computation speed is also relatively fast, which can basically meet the real-time requirements. It has done a lot of video sequence detection, and the effect is quite satisfactory;

 

9: Statistical Average Method

 

10: temporal median filter (median filter)

Automatic congestion detection system for underground Platform

Detecting Moving Objects, Ghost, and shadows in Video Streams

The statistical mean method and the median filter method are used to implement the statistical mean method and test the algorithm. The application of the algorithm has great limitations, it can only be regarded as a supplement in theory;

 

11: W4 Method

W4.pdf

The W4 algorithm should be one of the earliest algorithms used for practical application. You can check the relevant information and I will not elaborate on it here;

 

12: intrinsic background

A Bayesian computer vision system for modeling human interactions

Bayesian framework

 

13: Kernel Density Estimation

Non-parametric model for background subtraction

The last step is the kernel density estimation algorithm. This algorithm should be a relatively robust algorithm that can solve many algorithm parameter settings problems. Without setting parameters, it should be a major advantage of the algorithm.

 

Sobs, color, vibe, sacon, and PDF can be used for in-depth understanding, especially block-based, region-based, and features-based in recent years), algorithms based on hierarchical classification or hierarchical trainer can be studied in depth.

 

14: Optical Flow Method (the speed is too slow, basically not needed)

The concept of the optical flow method is derived from the optical flow field. When the image of a moving object moves in a pattern on the surface, it is called the optical flow field, which is a two-dimensional velocity field. The optical flow method calculates the movement size and direction of each pixel point based on the continuous multi-frame image sequence, which reflects the change trend of the gray scale of each image point on the image. Advantage: no background modeling is required. When you cannot obtain any information about the scenario in advance, you can detect independent motion objects. The disadvantage is that the computation is complex, and special hardware support is often required, which is difficult to meet real-time requirements.

 

15: http://code.google.com/p/bgslibrary/ this website library contains a variety of background subtraction methods, can save a lot of time.

 

16: Evaluation of Background Subtraction Techniques for video surveillance

Comprehensive Evaluation articles show the performance of various algorithms.

 

17: Analysis of foreground extraction by Wang Xianrong's blog (including code)

Http://www.cnblogs.com/xrwang/archive/2010/02/21/ForegroundDetection.html

 

18: Frame Difference Method

| Frame (I)-frame (I-1) |> TH, the background is the previous image. The difference between each frame and the previous frame is calculated. The extraction effect is obviously related to the speed and Frame Rate of the moving foreground object (the frame rate refers to the number of images in a second ). With the extension, the selective Background Modeling Based on the statistical model is actually the mixed gaussian method. Advantage: fast and stable. The disadvantage is that there may be "holes" in an object. The holes are due to a large moving object, and there is a close coincidence between the two frames of the object, therefore, this part is cut by difference. (The frame difference method mainly includes two-frame difference, three-frame difference, and cumulative Frame Difference)

 

19: Fixed background

| Frame (I)-background (I) |> TH, because the background is a preset fixed image, four problems must be introduced here: illumination change, camera jitter, high-frequency oscillating background, interference from moving to static objects. Advantage: the computation is simple and easy to implement. The disadvantage is that the camera head must be absolutely static and not adapt to illumination changes.

 

20: pixel-based adaptive segmenter

Background segmentation with feedback: The pixel-based adaptive segmenter

Http://www.mmk.ei.tum.de /~ HOM/PBAs website

The (PBAs) detection algorithm is a pixel-based parameter-free model. It combines the advantages of the sacon and vibe algorithms and is improved based on these two algorithms;

Main innovations:

(1) introduce the concept of control theory to make the foreground judgment threshold and the background model update rate adaptive and change with the complexity of the background.

(2) Introduce a measurement method of background complexity, and adjust the foreground judgment threshold and background model update rate based on the background complexity.

 

Foreground Target Detection 1 (Summary)