This paper presents a new P300 paradigm for
brain computer interface. Visual stimuli consisting of 8 arrows
randomly intensified are used for direction target selection for
wheelchair steering. The classification is based on a Bayesian
approach that uses prior statistical knowledge of target and
non-target components. Recorded brain activity from several
channels is combined with a Bayesian sensor fusion and then
events are grouped to improve event detection.
The system has an adaptive performance that adapts to user
and P300 pattern quality. The classification algorithms were
obtained offline from training and then validated offline and
online. The system achieved a transfer rate of 7 commands/min
with 95% false positive classification accuracy.