Différences

Ci-dessous, les différences entre deux révisions de la page.

--- supcomp_project [2009/12/09 16:38] – tigli
+++ supcomp_project [2009/12/09 17:07] (Version actuelle) – tigli
@@ Ligne 20: / Ligne 20: @@
 ==== Data format ====
+  * <listevent> ::= <event> | <event> <listevent>
   * event ::= <nom> <position 2D>
@@ Ligne 32: / Ligne 34: @@
 b : en bas à droite
+  * <listactor> ::= <actor> | <actor> <listactor>
   * actor ::= <Id> <Type> <Boundary box 3D> <Boundary box 2D> <Trajectory 3D> <CurrentPosition 3D> <Classification Properties>
@@ Ligne 58: / Ligne 63: @@
+==== Scenario ====
+The use of one or several cameras can recognize a wide range of activities for equipment less expensive.
+Unfortunatelly these devices are sensitive. They must be fixed and they must to be well calibrated with their associated recognition algorithms.
+The temptation is great to add low cost sensors in the user environment to have more robust detections.
+Nevertheless, this approach drastically increases the combinatorics of possible configurations.
+Also some equipment on moving objects allows to such configurations to vary over time.
+For example, a simple pressure sensor on the cushions of a chair can detect the presence of a seated person. Nevertheless, this chair can move and thus change the inferred position for the seated person.
+A carpet with a pressure sensor as presence detector, bought yesterday, will equally increment the user environment and the capacity of the activity recognition system.
+The recognition system activity should automatically adapt to the appearance / disappearance of new sensors.
+It should still be able to integrate the volley of new measures and merge them for a more robust activity  recognition (with uncertainties evaluation)