Demo IPOL: ViBe: a fast background subtraction algorithm

This repository constitutes the implementation of a demo IPOL of the popular background subtraction algorithm ViBe.

Requirements:

ffmpeg

Compilation :

The Makefile compiles the original source code of ViBe and combines it with the demo implementation so that different parameters can be easily tested. To compile, simply run:

make

Usage :

The algorithm can be run on a video file directly via the bash script or on a sequence of frames via de ./vibe command. Four different parameters are supported:

• -s [numberOfSamples]: The background reference model of ViBe consists in a collection of past, observed values for each pixel This parameter defines the number of stored values per pixel
• -r [matchingThreshold]: Threshold that regulates the classification of pixels into foreground or background, comparing them with the ones in the background model. A lower value makes the method less prone to false positives, while higher values will make it less prone to false negatives
• -c [matchingNumber]: Minimum number of values in the background model that need to be closer than the threshold to the oberved value, for it to be considered a background pixel.
• -uf [updateFactor]: Update factor to control the model update speed. For an update factor of 16, each pixel value classified as backgroud has one chance in 16 to be included in the background model.

Running with bash script :

The bash provided script takes a video input file, extracts its frames and applies ViBe on them. Another directory named masks/ is generated with the output masks from the algorithm. To run the bash scriptm run:

bash run.sh [path to input file] [numberOfSamples] [matchingThreshold] [matchingNumber] [updateFactor]

The output will consist of the following:

• imdir/: directory containing the input frames
• masks/: directory containint the output binary masks
• input.mov: video file of the input at a lower frame rate (this is done for demo visualization purposes)
• output.mov: video file of the output binary masks at the same frame rate as input.mov
• masks.zip: zip file containing the output frames (this is done for demo usability)

Running with the vibe command:

The algorithm can also be used using the /vibe command after compilation, which takes a set of RGB frames as input. To run the algorithm, you can simply do:

vibe [options] image1 image2 [image3 ... imageN]

For a directory imdir/ containing a set of frames as .png files, an example command to run the method is:

vibe -s 20 -r 20 -c 2 -uf 16 imdir/*png

This will create the output binary masks in the same directory of the provided input frames.