Ziyou's profile - activity

An image has two dimensions, thus have two focal length values.

1. Large scale and rotation is difficult for OF algorithms like the LK or Farneback. However, you may use previous estimation to warp current images toward the template image or the other way around.
2. I tested the speed of the LK and Farneback OF method in OpenCV. For calculating a dense flow field, the Farneback approach is in fact much faster than the LK algorithm. The reason that the LK is faster is because a sparse flow is usually computed.

Hi, guys Do you know how to calculate a sparse flow field using the calcOpticalFlowFarneback function in OpenCV? The Lucas Kanade algorithm has such an option where you can provide the sparse feature points to track. Thanks.

flow.at<vec2f>(i,j)[0] means x component of the flow at the position (i,j); flow.at<vec2f>(i,j)[1] means y component of the flow at the position (i,j);