Our online experiments are run on our mobile platform, namely Rackham and Jido. The different module are developped under GenoM architecture enabling module cooperation and data exchange

The following videos show the first results about Eigenface based face recognition.

Finding the person (Sylvain) among three persons in the image.		Akin, Brice and Thierry are the three persons known by the robot (its tutors), all the others are considered as unknown.

The following videos involve particle filtering visual tracking and face recognition. The tracking faults are detected thanks to the face recognition process.

Eigenface based tracking. The tracker only locks on the known person (black shirt).		Following the right target (blue shirt in the first, red shirt in the last) in spite of the disturbance (people crossing, partial occlusions, camera moving).			The robot focusses its attention on the first known person (Thierry) and follows him while he is in its field of view. Then it switches on the other known person : Akin.

Hereafter are presented some videos about visual 3D tracking. They involve blob triangulation for initialization, 3D estimation, and image based measurement.

Human 3D tracking. Dealing with disapearance, occlusions, target not gazing the camera.

The video presented hereafter shows offline evaluation of heterogenous data fusion

The four images in the video respectively represent:
- Top-Left : original image
- Top-Right : saliency map used for importance sampling
- Bottom-Left: particle cloud state after importance sampling and weight update
- Bottom-Right : particle cloud state after particle resampling and MMSE estimate

The videos presented hereafter show the first live experiments about heterogenous sensor data fusion within the particle filtering framework.

Fusion of visual cues and RFID detection. Camera servoing on visual template and RFID detection when visual template is absent.

Scenario : User kidnapping.			Scenario : Robot following.

Rackham : the Tour-Guide robot

Rackham is motionless, and looks at the room entrance. It fires its long-range monitoring modality, so as to detect and track any person entering the area. A group of two persons comes in. Their aim is to be led by Rackham to another place. As the group gets closer to the robot, the latter switches first to mid-range human tracking modality and then to the short-range head tracking modality, witch are sucessively more suited to the associated relative H/R distance. One person selects on the touchscreen ROBOT_HRP2 as destination. Then, the robot plans a path to ROBOT_HRP2 and asks the persons to stay behind it. The presence table ensure that at least one of the known persons is here and the robot starts the plan. During its motion, the robot switches to the upper human body tracking modality witch is more suited to the relative H/R distance and more suited when the robot moves. Even when an unknown person cross the field of view, the robot continues tracking one person of the group. When the robot loses its target, it suspends its plan to wait for one person of the group and recovers the mission.