Method for processing motion image to determine optical flow and system for same

An image processing system in a motion image measuring apparatus which measures movement of an object by processing a sequence of motion images of the object, said sequence of motion images being captured by an image pick-up camera and said sequence of motion images having a plurality of brightness patterns, the system comprising:

• spatiotemporal filtering differentiator means for filtering and differentiating the plurality of brightness patterns, the spatiotemporal filtering differentiator means having a plurality of differing initial phases for obtaining a set of local differential values grouped by the plurality of differing initial phases;
• calculating means for calculating local brightness gradients based on the local differential values;
• and
• measuring means for providing a measurement of movement of said object in accordance with the local brightness gradients.

An image processing method in a motion image measuring apparatus which measures movement of an object by processing a temporal sequence of motion images of the object in order to determine an optical flow in terms of a brightness gradient at a spatiotemporal local region in a frame of the temporal sequence of motion images, each of said motion images comprising an original brightness pattern, the original brightness pattern comprising:

• geometric real local areas wherein each geometric real local area is defined in a real space in a spatiotemporal frame,
• local measures of brightness wherein each local measure of brightness is associated with one of the geometric real local areas, and
• noise components wherein each noise component is associated with one of the geometric real local areas,
• the method comprising the steps of:
• an initial step of capturing the temporal sequence of motion images of said object in an image pick-up camera;
• a first step comprising substeps (a) through (j), wherein substeps (a) through (d) occur separate from, substeps (e) through (j), substeps (a) through (j) being defined as follows:
• (a) removing the noise component associated with each geometric real local area of a current motion image from the temporal sequence of motion images, by filtering the geometric real local area through a plurality of anisotropic spatiotemporal filters, wherein each the anisotropic spatiotemporal filters has a plurality of parameters, and wherein one of the parameters is an initial phase shifting parameter;
• (b) obtaining, for each geometric real local area, a plurality of first contrasty local areas, each of the plurality of first contrasty local areas being identifiable by the parameters corresponding to one of the plurality of anisotropic spatiotemporal filters;
• (c) executing a convolution of the local measure of brightness associated with each geometric real local area to the first contrasty local areas;
• and
• (d) obtaining a plurality of identifiable local brightness estimates from the convolution of substep (c);
• (e) removing the noise component associated with each geometric real local area of a current image from said temporal sequence of motion images by filtering the geometric real local areas through a plurality of anisotropic spatial filters, said plurality of anisotropic spatial filters having a plurality of parameters in a one-to-one correspondence to the plurality of parameters of the anisotropic spatiotemporal filters;
• (f) obtaining a plurality of second contrasty local areas, each second contrasty local area being identifiable by the parameters corresponding to one of the anisotropic spatial filters;
• and
• (g) determining a difference between each second contrasty local area and a corresponding second contrasty local area of a neighboring motion image in the temporal sequence of motion images;
• (h) obtaining a plurality of geometric local differences;
• (i) executing a convolution of a local measure of brightness associated with each geometric real local area to the geometric local differences;
• and
• (j) obtaining a plurality of identifiable local difference estimates of brightness from the convolution of substep (i);
• a second step comprising the substeps of:
• determining a ratio of each identifiable local difference estimate to a corresponding one of the identifiable local brightness estimates;
• obtaining for each of the phase shifting parameters a plurality of first local brightness differential values;
• determining a total sum of the local brightness differential values from all of the phase shifting parameters;
• and
• obtaining for each parameter a plurality of second local brightness differential values;
• a third step comprising the substeps of:
• determining a local brightness gradient for each geometric real local area in the temporal sequence of motion images;
• and
• storing the local brightness gradient in an accessible manner;
• and
• a fourth step comprising the substeps of:
• reading the local brightness gradient of each of the geometric real local areas in a spatiotemporal vicinity of the local region;
• and
• calculating the brightness gradient at the spatiotemporal local region based on the local brightness gradients read.