Background modeling Technology (II): Bgslibrary Framework, background modeling of 37 algorithms performance analysis, background modeling technology challenges

Background modeling Technology (II): Bgslibrary Framework, background modeling of 37 algorithms performance analysis, background modeling technology challenges

1. bgslibrary software download based on MFC

: http://download.csdn.net/detail/frd2009041510/8691475

The software platform includes 37 background modeling algorithms, which can display the input video/image, background image based on background modeling and background modeling, and also show the computational complexity of each algorithm. Also, the test can be a video, a picture sequence, and a camera input video. Its interface is as follows:

2, the Bgslibrary framework

Bgslibrary's software framework, as shown, is based on the C + + platform, as detailed source code will be analyzed in the next blog.

3. Performance comparison of various background modeling algorithms in Bgslibrary

The performance of various background modeling algorithms is compared in three aspects of CPU utilization, required memory and running time.

4. Problems that can be solved by various background modeling algorithms

Background modeling techniques Encounter the following challenges:

(1), noise Picture: This situation basically belongs to the network camera or compressed video image caused by low-quality image/video source;

(2), camera jitter: Under some conditions, the wind will cause the camera to shake before and after;

(3), Camera automatic adjustment: At present, many cameras have automatic control functions, such as lighting control, white balance and zoom out functions;

(4), light changes: This is the most common, such as the light on the light-off light changes;

(5), Bootstrap phenomenon: in a period of time, background modeling technology can not get the background image of the situation;

(6), disguise: foreground and background image pixel value is very similar, resulting in the foreground and background can not distinguish;

(7), foreground hole: When using background subtraction, some areas within the foreground may be mistaken for background;

(8), moving background object: Some objects in the background may also move, these objects cannot be regarded as part of the foreground;

(9), enter the prospect goal: A new moving object enters, should be able to be detected in time;

(10), dynamic background: such as small target jitter, leaf jitter, water ripples;

(11), start the goal of the movement: when the target in the scene begins to move, it itself and the new related parts should be detected;

(12), the prospect of entering the dormant target: some prospects of the target into the scene after the static situation;

(13), Shadow: Can detect the shadow of the moving object and the original shadow of the background area.

Faced with these challenges, the list of issues that some background modeling techniques can solve is as follows:

Background modeling Technology (II): Bgslibrary Framework, background modeling of 37 algorithms performance analysis, background modeling technology challenges