CODEWITHSUNDEEP
pythonopencv
Harsh Wardhan
Harsh Wardhan

Reputation: 2148

Range of output values in calcOpticalFlowFarneback function in Python OpenCV

I am calculating the optical flow for a video using

flow = cv2.calcOpticalFlowFarneback(prvs,next, None, 0.5, 3, 15, 3, 5, 1.2, 0)

The input resolution is 320x240. I computed some basic stats for the flow data received from the function with this code

arr1 = np.load(file_path)
y = arr1[:,:,0]
x = arr1[:,:,1]

if (y_min > y.min()):
    y_min = y.min()

if (y_max < y.max()):
    y_max = y.max()

and I got the following values:

y:

    min                max                mean                std_dev
-838.59191895        850.21942139        0.01124349        4.41635523



x:

    min                max                mean                std_dev
-58.26990128        73.48989105        0.00110086        2.47226620

I noticed that for y coordinates the min and max values far exceed the input dimensions, i.e., 320x240. Can anybody point out the reason for this observation? I'm unable to figure out the valid range of values expected from the cv2.calcOpticalFlowFarneback function.

Upvotes: 0

Views: 682

Answers (1)

Tobias Senst
Tobias Senst

Reputation: 2830

Theoretically there is no minimum and maximum bound in the optical flow estimates, since there is no search range as by block matching methods. The motion vector is derived from the image gradients and the optical flow equation and can point outside the image or video bounds. The high errors might be gross-outliers of flow estimation which is totaly normal and may be caused if the content in the images change dramatically due to illumination changes, shadows, motion blur, coding artifacts etc.

Upvotes: 2

Related Questions