Please enable JavaScript.
Coggle requires JavaScript to display documents.
Computational Vision 2 (Motion (Matching points of interest (Discreteness …
Computational Vision 2
Motion
Motion Correspondence - Picking "interesting" points such as corners and tracking those. Can then describe exactly how an object is moving.
-
-
-
Detecting a change - If camera moved, then difference is calculated but the subtraction of a feature through 2 panes.
Moravec Operator - Defines points of interest by moving an nxn mask around and finding where intensity changes greatly. This detects corners well. Noise must be removed first
-
-
Object Recognition
-
Model database - list of objects which you are interested in. Information stored can be images, dimensions, colour, texture, speed etc. Needs to be well organised for computation
-
Hypothesizer - Looks at all of the features and determines the likelihood that it is an object in the model database. Eliminates useless features
Hypothesis verifier - Actually assigns the label to the remaining features. Uses all available information such as likelihood of this object being next to this object etc.
-
Model Object Recognition
Marr's Approach
-
-
-
-
-
-
-
-
Primal Sketch (Intensity changes and likelihood of depth change) to 2.5D sketch (adding info together to get a general idea of the world) to 3D
General Shapes - Algorithm may look for a cylinder object as a human, and then break it down into smaller parts to match head, arm, hand etc.
Geons - Small library of 3D geometric shapes that are used to match to objects. Eg. a mug is a cylander and a curved cylander
Statistical Approach - Storing characteristics of an object so it can be recognised. Very view dependant
-
Constellation model - Where features of an object are grouped all to one another. May not be ideal if one of the features is missing due to appearance