100 likes | 250 Views
Computer Vision REU Week 3. Adam Kavanaugh. Lucas Kanade Method. Algorithm Overview Create Gaussian Kernels Convolute both images with dx and dy Gaussian Kernel Take the average of the resulting convolutions between both images in x and y direction
E N D
Computer Vision REUWeek 3 Adam Kavanaugh
Lucas Kanade Method • Algorithm Overview • Create Gaussian Kernels • Convolute both images with dx and dy Gaussian Kernel • Take the average of the resulting convolutions between both images in x and y direction • Convolute the images with the Gaussian Kernel • Take the difference of the Gaussians • Create the vector field of the optical flow
Differences in Implementations • For Simplicity, in my conversion, I defined any matrix which is singular to have the vector position <0,0> for the field. • In MATLAB, the function pinv() is used on the matrix to be inverted. • This function is a Moore-Penrose pseudo inverse • Computed using singular value decomposition
Original Testing Frames • Provided by Edward. Frame 12 Frame 13
Gaussian Pyramid • Stripped down a larger version of the code • Written in C • Handles only .pgm files
Topics of Interest • Applications involving optical flow • Tracking • Obstacle Detection • Object recognition • Image Mosaicing • Content based retrieval