Background Modeling-Gaussian Mixture Model

ReferencesAdaptive Background Mixture Models for real-time tracking cvpr 1999

Assuming that the prerequisite pixel shows the probability model of Gaussian Model on the timeline

The Gaussian mixture model uses 3 ~ Five Gaussian models are used to characterize the features of each pixel in the image. After a new frame is acquired, the Gaussian mixture model is updated.

Match each pixel in the current image with the Gaussian mixture model. If the model is successful, it is determined that the point is the background point, otherwise it is the foreground point.

The Gaussian mixture model selects an update strategy based on different matching conditions. There are two learning factors used for mean and variance update learning.

Improvement Method:

1) adopt different learning mechanisms to update variance and mean;

2) Adjust K size based on the complexity of the background to reduce the computing complexity;

3) select other features to establish a Gaussian mixture model, such as texture features, texture and color features.

Main advantages: it can adapt to the influence of light gradient on the background

Main disadvantages: high computing complexity. Secondly, false detection caused by camera jitter on leaves cannot be avoided. When the foreground is similar to the background color, foreground errors will be processed as background errors.

The update formula is as follows:

DetailsCodeImplementation, refer to the implementation in opencv,Bgfg_gaussmix.cpp

--