Cognitively-Inspired Architectures for Human Motion Understanding

reportActive / Technical Report | Accesssion Number: AD1228741 | Open PDF

Abstract:

In the last decades, modelling and understanding human motion from videos has gained an increasing importance in several application domains, including Human-Machine Interaction, gaming, assisted living and robotics. Although the significant advances of the last years, where as in other domains deep learning techniques has gained momentum, the tasks remain among the most challenging, for the intrinsic complexity of dynamic information, and still a lot of work needs to be done to approaching human performance. The biological perceptual systems remain the gold standard for efficient, flexible, and accurate performance across a wide range of complex real-world tasks. A natural inspiration for computational models are thus the mechanisms underlying human motion perception, and the knowledge derived from the Cognitive and Neuro Science fields. Previous works demonstrated the effectiveness of biologically-inspired visual features for object or action recognition , while examples of cognitively-inspired architectures are less present.

Security Markings

DOCUMENT & CONTEXTUAL SUMMARY

Distribution Code:
A - Approved For Public Release
Distribution Statement: Public Release.
Copyright: Not Copyrighted

RECORD

Collection: TRECMS
Subject Terms