Motion Segmentation
Welcome to Motion Segmentation a Computational Photography project page.
Welcome to Motion Segmentation a Computational Photography project page.
As the name may imply, motion segmentation divides a video into different regions based on detected motion. That is, neighborhoods of similar motion can be grouped together in a single layer, allowing for a dynamic scene to be broken into components of individual motion. The utility of this operation relies on the fact that objects in the real world tend to undergo smooth movements as coherent wholes. Thus, detecting these regions of similar motion allows us to detect moving objects in the scene.
Optical flow algorithms perceive any change in color or intensity as motion, and this introduces several potential sources of error. To reduce this we need to make an assumptions called the optical flow constraint .
Our implementation best detects rigid motion. A Rigid Motion preserves the size and shape of an object, and includes rotations or translations. As the dice demonstrate rigid motion provides uniform colored vectors.
Subjects that do not undergo rigid motion are not accurately characterized as a single region of motion. Hence they are often segmented into multiple layers. Fire displays fantastical colors due to the confused vectors generated
For internal compositing all manipulations occur inside MATLAB scripts. We layer mask a foreground moving object onto another video or static image.
For external compositing we write a green colored background to the isolated motion footage to act as an procedurally generated green screen. This footage is used with the chroma keying functionality of a dedicated video editor.
Andrew Chase, Bryce Sprecher, and James Hoffmire