Experiments on FROG data Sample Clauses
Experiments on FROG data. In FROG project, our framework was used for detection of social attitudes like agreeing and disagreeing. Detection is based on the state of the art in cognitive sciences and based on morphological and temporal correlations between relevant visual cues, including facial gestures like frowns and smiles as well as head gestures like tilts and nods. As features, the facial point landmarks as well as the head pose were used producing very satisfactory results. The method for facial landmark detection as well as for face pose estimation was developed for FROG and it was presented in deliverable 3.
Experiments on FROG data. In FROG project, the robot only has to remember a limited number of faces belong- ing to the visitors interacted with it since the beginning of the tour. These faces are stored in a non-persistent in-memory database called the gallery. For each frame, the robot compares every visible human face with every face stored in this gallery. For our face recognition experiments in FROG data, we have collected by about 4 hours of real FROG video recordings. In each video, up to 140 different persons were depicted. Our face recognition experiments show that the 99.8% of the human faces can be correctly recognised, which means that our method performs extremely well in outdoor environ- ments. Figure 2.9 depicts some real visual FROG face recognition results.