processors#

Individual transformation operations for video frames.

To be used with the perceptivo.video.pupil.PupilExtractor subclasses

class perceptivo.video.processors.Processor(**kwargs)#

Bases: perceptivo.root.Perceptivo_Object

Individual processing stage, can be added together to make a processing chain

__add__ method based on autopilot.transform.transforms.Transform

property parent: Optional[perceptivo.video.processors.Processor]#

If this Transform is in a chain of transforms, the transform that precedes it

Returns

Transform, None if no parent.

abstract process(input: Union[perceptivo.types.video.Frame, perceptivo.types.units.Ellipse]) → Union[perceptivo.types.video.Frame, perceptivo.types.units.Ellipse]#

Process a frame!

Typically you want a chain of processors to end up outputting an Ellipse, but this is not enforced

Returns:

__add__(other)#

Add another Transformation in the chain to make a processing pipeline :Parameters: other (Transformation) – The transformation to be chained

class perceptivo.video.processors.Canny(blur_sigma: float = 1, low_thresh: float = 0.2, high_thresh: float = 0.5)#

Bases: perceptivo.video.processors.Processor

Canny edge detection.

Slight modification of skimage.feature.canny(), but using opencv and Scharr kernel rather than sobel for better orientation invariance, and also using the eigenvectors of the structure tensor rather than the simple hypotenuse

The source for this class is really blippy because it is optimized for speed!

Variables

  • blur_sigma (float) – Amount of blurring to use in the initial smoothing step

  • low_thresh (float) – Low threshold for canny edge detection

  • high_thresh (float) – High threshold for canny edge detection

References

process(frame: perceptivo.types.video.Frame) → perceptivo.types.video.Frame#

Process a frame!

Typically you want a chain of processors to end up outputting an Ellipse, but this is not enforced

Returns:

class perceptivo.video.processors.Haar_Tracker(tracker_type: str = 'eye', min_neighbors: int = 20, adaptive_neighbors: bool = True, **kwargs)#

Bases: perceptivo.video.processors.Processor

Download and use a haar cascade to track.

Many trained cascade .xml files are available at https://github.com/opencv/opencv/tree/master/data/haarcascades

References

XML_URLS = {'eye': 'https://raw.githubusercontent.com/opencv/opencv/415a42f327104653604fc99314eb215cd938d6d7/data/haarcascades/haarcascade_eye.xml', 'face_default': 'https://raw.githubusercontent.com/opencv/opencv/415a42f327104653604fc99314eb215cd938d6d7/data/haarcascades/haarcascade_frontalface_default.xml'}#
property url: str#
property filename: pathlib.Path#
process(frame: perceptivo.types.video.Frame) → Tuple[List[tuple], List[int]]#
Parameters

frame (Frame)

Returns

list of tuples corresponding to (x,y,width,height)

class perceptivo.video.processors.Filtered_Hough(radii=(15, 30, 100), max_considered=3, peaks_kwargs: Optional[dict] = None)#

Bases: perceptivo.video.processors.Processor

A hough transform to detect circles, returning the one that bounds the darkest area in the image

process(edges: numpy.ndarray)#

Frame to process, along with edges from canny edge detection

class perceptivo.video.processors.Filter_Circles(prior_bias=0.5)#

Bases: perceptivo.video.processors.Processor

Filter Circles!

Parameters

prior_bias (float) – how strongly to weight the similarity to the prior circle, if given

__init__(prior_bias=0.5)#
Parameters

prior_bias (float) – how strongly to weight the similarity to the prior circle, if given

process(frame: perceptivo.types.video.Frame, circles, prev_eye=None)#

Process a frame!

Typically you want a chain of processors to end up outputting an Ellipse, but this is not enforced

Returns:

perceptivo.video.processors.circle_to_mask(frame, ix, iy, rad)#