fgGaussian Class Reference

Single Gaussian target/foreground model with diagonal covariance. More...

Inheritance diagram for fgGaussian:

Collaboration diagram for fgGaussian:

Public Member Functions
def	__init__ (self, bgCfg=None, processor=None, fgMod=None)
	Constructor for single Gaussian model background detector. More...
def	adapt (self)
	Update the Gaussian model based on recent measurement. More...
def	correct (self)
	Generate a correction to the model. More...
def	detect (self, I, M=None)
	Given a new measurement, apply the detection pipeline. More...
def	displayState (self)
def	displayState (self)
def	emptyDebug (self)
	Return empty debug state information. More...
def	emptyState (self)
	Return empty state. More...
def	estimateFromData (self, theData)
	Use given data to estimate Gaussia foreground model. More...
def	estimateFromMaskRGB (self, theMask, theImage)
	Extract mask region pixel data from image for Gaussia model estimation. More...
def	get (self)
def	getDebug (self)
	Return current/latest debug state information. More...
def	getState (self)
	Return current/latest state. More...
def	info (self)
	Provide information about the current class implementation. More...
def	measure (self, I, M=None)
	Takes image and generates the detection result. More...
def	predict (self)
	Predictive model of measurement. More...
def	process (self, I)
	Given a new measurement, apply entire FG modeling pipeline. More...
def	refineFromRGBDStream (self, theStream, incVis=False)
def	refineFromStreamRGB (self, theStream, incVis=False)
	Given an RGB stream, run the estimation process with adaptation on to improve color model. More...
def	refineFromStreamRGBD (self, theStream, incVis=False)
def	saveConfig (self, outFile)
	Save current instance to a configuration file. More...
def	saveTo (self, fPtr)
	Empty method for saving internal information to HDF5 file. More...
def	set (self)
def	testOnRGBDStream (self, theStream, incVis=True)
def	updateFromData (self, theData, alpha=None)
	Use given data to update Gaussia foreground model based on learning rate. More...
def	updateFromMaskRGB (self, theMask, theImage, alpha=None)
	Extract mask region pixel data from image for Gaussia model update/adaptation. More...
Public Member Functions inherited from fgImage
def	__init__ (self, processor=None)
Public Member Functions inherited from inImage
def	measure (self, I)
	Generate detection measurements from image input. More...
Public Member Functions inherited from Base
def	__init__ (self)
	Instantiate a detector Base activity class object. More...
def	detect (self, signal)
	Run detection only processing pipeline (no adaptation). More...
def	save (self, fileName)
	Outer method for saving to a file given as a string. More...

Static Public Member Functions
def	buildFromCfg (theConfig, processor=None, fgMod=None)
	Instantiate from stored configuration file (YAML). More...
def	load (fileName)
	Load Gaussian instance specification from HDF5 file. More...
def	loadFrom (fPtr)
	Specialzed load function to parse HDF5 data from file pointer. More...
def	loadFromYAML (fileName)
	Instantiate from stored configuration file (YAML). More...
Static Public Member Functions inherited from Base
def	load (fileName, relpath=None)
	Outer method for loading file given as a string (with path). More...

Public Attributes
	bgI
	config
	errI
	fgI
	improcessor
	imsize
	maxE
	measI
	measM
	mu
	nrmE
	sigma
	sqeI
Public Attributes inherited from inImage
	Ip
	processor
Public Attributes inherited from Base
	x
	Detection state. More...

Detailed Description

Single Gaussian target/foreground model with diagonal covariance.

A similar implementation exists in targetSG, based on a different operating paradigm that decorrelates or whitens the input image data, which is effectively a linear transformation of the image data. While the implementation is most likely better than this one, simplicity has its own value.

No doubt this implementation exists in some form within the OpenCV or BGS libraries, but getting a clean, simple interface from these libraries is actually not as easy as implementing from existing Matlab code. Plus, it permits some customization that the library implementations may not have.

For how to configure the input, please see CfgSGT. Likewise for model overriding, please see SGMdebug.

Note

A note on the improcessor. If the basic version is used, then it performs pre-processing. If a triple version is used, then the mid-processor will perform operations on the detected part rather than the default operations. The mid-processor can be used to test out different options for cleaning up the binary data.

Todo:

Eventually should be translated to OpenCV style code by repurposing their code to get CUDA and OpenCL implementations for speed purposes. Even their C code is fairly zippy relative to python.

Todo:

Redo so that inherits from appearance or some kind of fgmodel class.

Constructor & Destructor Documentation

__init__()

def __init__	(		self,
			bgCfg = None,
			processor = None,
			fgMod = None
	)

Constructor for single Gaussian model background detector.

Parameters

[in]	bgMod	The model or parameters for the detector.
[in]	bgCfg	The model or parameters for the detector.

Member Function Documentation

adapt()

def adapt

(

self

)

Update the Gaussian model based on recent measurement.

In this case, the mean and the variance are updated. Depending on the run-time options, all means/variances will be updated or only those classified as background. The latter avoids adapting to foreground elements while still permitting slow change of the background model.

Usually, during the model estimation phase (assuming an empty scene with background elements only) adaptation of all pixels should occur. During deployment, if adaptation is to be performed, then it is usually best to not apply model updating to foreground elements, which are interpreted as fast change elements of the scene.

Reimplemented from Base.

buildFromCfg()

def buildFromCfg (   theConfig,   processor = None,   fgMod = None  )

static

Instantiate from stored configuration file (YAML).

correct()

def correct

(

self

)

Generate a correction to the model.

In standard schemes, there are no corrections to the estimates. The classification result is presumed to be correct. Corrections would imply some sort of temporal regularization. Spatial regularization is usually done through image-based mid-processing.

Reimplemented from Base.

detect()

def detect	(		self,
			I,
			M = None
	)

Given a new measurement, apply the detection pipeline.

This only goes so far as to process the image and generate the detection, with correction. There is no adaptation. It should be run separately if desired.

displayState() [1/2]

def displayState

(

self

)

displayState() [2/2]

def displayState

(

self

)

emptyDebug()

def emptyDebug

(

self

)

Return empty debug state information.

Useful if contents needed beforehand.

Reimplemented from Base.

emptyState()

def emptyState

(

self

)

Return empty state.

Useful if contents needed beforehand.

Reimplemented from Base.

estimateFromData()

def estimateFromData	(		self,
			theData
	)

Use given data to estimate Gaussia foreground model.

Parameters

[in]

theData

Column-vectors of target data.

Todo:

Establish if should be row or column.

estimateFromMaskRGB()

def estimateFromMaskRGB	(		self,
			theMask,
			theImage
	)

Extract mask region pixel data from image for Gaussia model estimation.

Parameters

[in]	theMask	Regions of interest w/binary true values.
[in]	theImage	Source image to get model data from.

get()

def get

(

self

)

getDebug()

def getDebug

(

self

)

Return current/latest debug state information.

Usually the debug state consists of internally computed information that is useful for debugging purposes and can help to isolate problems within the implemented class or with downstream processing that may rely on assumptions built into this implemented class.

Reimplemented from Base.

getState()

def getState

(

self

)

Return current/latest state.

Reimplemented from Base.

info()

def info

(

self

)

Provide information about the current class implementation.

Exists for reproducibility purposes. Usually stores the factory information used to build the current class instance.

Reimplemented from inImage.

load()

def load (   fileName )

static

Load Gaussian instance specification from HDF5 file.

loadFrom()

def loadFrom (   fPtr )

static

Specialzed load function to parse HDF5 data from file pointer.

Reimplemented from Base.

loadFromYAML()

def loadFromYAML (   fileName )

static

Instantiate from stored configuration file (YAML).

measure()

def measure	(		self,
			I,
			M = None
	)

Takes image and generates the detection result.

Parameters

[in]

I

Image to process.

predict()

def predict

(

self

)

Predictive model of measurement.

In standard schemes, the expectation is that the background model is static (a constant state model). Thus, the default prediction is no update.

Reimplemented from Base.

process()

def process	(		self,
			I
	)

Given a new measurement, apply entire FG modeling pipeline.

Parameters

[in]

I

New image measurement.

Reimplemented from Base.

refineFromRGBDStream()

def refineFromRGBDStream	(		self,
			theStream,
			incVis = False
	)

refineFromStreamRGB()

def refineFromStreamRGB	(		self,
			theStream,
			incVis = False
	)

Given an RGB stream, run the estimation process with adaptation on to improve color model.

Parameters

[in]	theStream	RGB stream.
[in]	incVis	Include visualization of refinement processing [False]

refineFromStreamRGBD()

def refineFromStreamRGBD	(		self,
			theStream,
			incVis = False
	)

saveConfig()

def saveConfig	(		self,
			outFile
	)

Save current instance to a configuration file.

saveTo()

def saveTo	(		self,
			fPtr
	)

Empty method for saving internal information to HDF5 file.

Save data to given HDF5 pointer. Puts in root.

Reimplemented from Base.

set()

def set

(

self

)

testOnRGBDStream()

def testOnRGBDStream	(		self,
			theStream,
			incVis = True
	)

updateFromData()

def updateFromData	(		self,
			theData,
			alpha = None
	)

Use given data to update Gaussia foreground model based on learning rate.

Parameters

[in]	theData	Column-vectors of target data.
[in]	alpha	[None] Learning rate override: set for value other than internal one.

Todo:

Establish if should be row or column.

updateFromMaskRGB()

def updateFromMaskRGB	(		self,
			theMask,
			theImage,
			alpha = None
	)

Extract mask region pixel data from image for Gaussia model update/adaptation.

Parameters

[in]	theMask	Regions of interest w/binary true values.
[in]	theImage	Source image to get model data from.
[in]	alpha	[None] Learning rate override: set for value other than internal one.

Member Data Documentation

bgI

bgI

config

config

errI

errI

fgI

fgI

improcessor

improcessor

imsize

imsize

maxE

maxE

measI

measI

measM

measM

mu

mu

nrmE

nrmE

sigma

sigma

sqeI

sqeI

---

The documentation for this class was generated from the following file:

• /home/pvela/python/detector/detector/fgmodel/Gaussian.py