| 摘要 |
A video object identification system used to identify an object, to locate its position in the view, and to determine its angular orientation. It digitizes a field of image view and subdivides the viewed area into a grid pattern of small cells. It then encodes the angle of a tangent line segment within each cell (when present) at the boundary of objects in the view. It determines rate-of-curvature of the boundary to develop a linear signature for the object. The system breaks the linear signature into segments that can be normalized to be constant regardless of the scale of the image. It then generates identification numbers to identify the object through a mathematics process. The system utilizes pipeline processing to the point where the results are applied to a microprocessor. The microprocessor analyzes a stored image field of view in encoded format to determine object perimeter cell locations (a chaining process). It then traces this perimeter addressing the stored image field to recover processed differential angles and to control the development of a linear signature sequence. The mathematics process performed creates a first number representing the length of straight segments and creates a second number representing the degree of curvature of curved segments. The combination of the two numbers provides a high probability of unique identification of an object. The numbers are compared with established numbers of a template for identification of an object. A unique feature number of an object is established for the template. This feature is located on the chained object by segment comparison and the relative angular orientation, compared to the template orientation, is then determined.
|
