I am trying to set parameters for GMG background subtractor algorithm in opencv for camera inputs from different kinds of environments. example, urban road, rural roads, small well lit room, big dimly lit hall, etc. I currently have access to only a few types of cameras hence I am not able to develop a good intuition about how these parameters are to be set with hit and trial. I am unable to have a good mathematical understanding of the paper. The default parameter values and definitions of parameters are as follows.
// Background prior: prior probability that each individual pixel is a background pixel.
backgroundPrior = 0.8
// Decision Threshold: value above which pixel is determined to be FG.
decisionThreshold = 0.8
// Learning Rate: It lies between 0.0 and 1.0. It determines how quickly features are "forgotten" from histograms.
learningRate = 0.025
// Max Features: total number of distinct colors to maintain in histogram.
maxFeatures = 64
// Quantization Levels: It is the number of discrete levels in each channel to be used in histograms.
quantizationLevels = 16
// Smoothing Radius: kernel radius used for morphological operations.
smoothingRadius = 7
I would like to set the parameters for each type of environment optimally.
