Kolloquium Wintersemester 2011/12
Die Dozenten der Informatik
Jun.-Prof. Dr. Björn Ommer, University of Heidelberg
spricht über
Recognition by Grouping Dependent Parts
| Datum: | Freitag, 9. März 2012
| | Zeit: | 14:00 Uhr
| | Ort: | Hörsaal III.03 im Landesbehördenhaus LBH, Friedrich-Ebert-Allee 144
|
ZusammenfassungA common approach to multi-scale, category-level object detection in cluttered scenes are voting methods. Despite the current popularity, Hough voting has two significant weaknesses: i) (semi-)local parts are independently casting their votes for the object hypothesis and ii) intrinsically global object characteristics like shape are assumed to be a mere sum of local parts. This assumption is against the fundamental conviction of Gestalt theory that the whole object is different from the sum of its parts and it violates the properties of widely used, semi-local features. It is however possible to actually utilize these dependencies by integrating shape-based contour grouping into the voting procedure. Therefore, mutually dependent parts are grouped while solving the correspondence problem jointly for all the dependent parts in a composition.
The compositional grouping of object parts can be extended to the parsing of complete scenes and videos and it provides a feasible approach to abnormality detection. Therefore, video frames are parsed by establishing a set of hypotheses that jointly explain all the foreground while, at the same time, trying to find normal training samples that explain the hypotheses. Consequently, a direct detection of abnormalities can be avoided. This is crucial since the class of all irregular objects and behaviors is infinite and thus no (or by far not enough) training samples are available.
(Vortrag auf Einladung der Dozenten der Informatik)
Letzte Änderung:
21. Februar 2012, 13:51:42
|
.
Wintersemester 2011/12