From Color Groups to Scene Interpretation

Nicolas Jouandeau (1) , P. Bonnin (2) , Vincent Hugel (2) , P. Blazevic (2)

1 LIASD - Laboratoire d'Informatique Avancée de Saint-Denis
2 LISV - Laboratoire d'Ingénierie des Systèmes de Versailles

In this paper we detail an algorithm that deals with objects and shape recognition based on color groups identification. The algorithm can dynamically adjust the unit set of pixels used to define color groups. Each color group is selected from the current scene. The algorithm is optimized to run faster than classical ones. Experiments were conducted on soccer images captured from real situations of the standard platform league (SPL). They also show the influence of camera parameters like contrast and gain. They lead to the conclusion that a set of rules could help to aggregate color groups and to identify scenes according to a classical predefined color look-up table (CLUT).

Domaines: Intelligence artificielle [cs.AI]

Nicolas Jouandeau, P. Bonnin, Vincent Hugel, P. Blazevic. From Color Groups to Scene Interpretation. 4th Workshop of Humanoïd Soccer Robots, Dec 2009, Paris, France. ⟨hal-02310200⟩