Bio-inspired computational model of perception for the detection of simple events from their visual features.

Event segmentation in temporal data streams is a key challenge in cognitive
architectures, with applications in robotics, video analytics, natural language
processing, and many other areas. This work presents an innovative event seg-
mentation model inspired by neuroscience and psychology, two branches of
cognitive science. The model emulates the temporal segmentation processes of the
human brain during visual perception. The model integrates perceptual processes
with hierarchical attention mechanisms to detect transitions between events in
dynamic environments. This integration enables the generation of an experience-
based knowledge structure that cognitive functions, such as memory, planning,
and decision-making, use to perform their tasks. To evaluate the model’s robust-
ness, a case study was conducted in a real-life environment. Continuous streams
of visual data obtained through a camera emulating eye behavior were trans-
formed into simple and meaningful events based on previous experience. The
results demonstrate the model’s ability to identify events with high accuracy,
even in dynamic and noisy conditions. This cognitive science–inspired approach
offers a promising framework for adaptive cognitive systems with potential appli-
cations in autonomous robotics and behavioral analysis. Our current research
explores the model’s generalization to other sensory domains and its integra-
tion into more complex cognitive systems.