Histograms of Optical Flow¶

Annotator to extract dense Optical Flow.

Uses the opencv Gunnar Farneback’s algorithm and represent it as a histogram of optical flow orientation and magnitude (HOFM).

class dvt.annotate.hofm.HOFMAnnotator(**kwargs)[source]¶

Bases: dvt.abstract.FrameAnnotator

Annotator to extract dense Optical Flow using the opencv Gunnar Farneback’s algorithm and represent it as a histogram of optical flow orientation and magnitude (HOFM).

The annotator will return the optical flow describing the motion in two subsequent frames as a HOFM feature.

freq¶

How often to perform the embedding. For example, setting the frequency to 2 will computer every other frame in the batch.

Type:	int

blocks¶

How many spatial blocks to divide the frame in, in each dimension. Default is 3, which results in 9 spatial blocks.

Type:	int

mag_buckets¶

List of bounds for magnitude buckets. Default is [0, 20, 40, 60, 80, 100].

Type:	list of ints

ang_buckets¶

List of bounds for angle buckets. Default is [0, 45, 90, 135, 180, 225, 270, 315, 360].

Type:	list of ints

frames¶

An optional list of frames to process. This should be a list of integers or a 1D numpy array of integers. If set to something other than None, the freq input is ignored.

Type:	array of ints

name¶

A description of the aggregator. Used as a key in the output data.

Type:	str

annotate(batch)[source]¶

Annotate the batch of frames with the optical flow annotator.

Parameters:	batch (FrameBatch) – A batch of images to annotate.
Returns:	A list of dictionaries containing the video name, frame, and the HOFM optical flow representation of length (len(blocks) * len(blocks) * len(mag_buckets) * len(ang_buckets))

name = 'hofm'