Human Activity Recognition System Based-on Sequential Logic Circuits and Statistical Models

Abstract

this research proposed the human activity
recognition system that described complete flow of processes from
lowest process (dealing with images) to highest process (recognize
human activity). We proposed human action recognition that
manage image sequence then recognize human action with simple
human model by model-based recognition technique. The
experimental result shows good accuracy which up to 93%
correctly recognized. We proposed the human activity process
with 3 methods that consecutive improved. All of those methods
can use the result of action recognition as inputs. First method is
FSM recognizer. The human model in Finite State Machine (FSM)
recognizer can be modeled by rational condition that make it easy
to understand and consume low computation cost but it hard to
define complex activity condition so it is unsuitable method for
complex activity. The second recognizer applied Hidden Markov
Model (HMM) for activity modeling. The HMM recognizer can
dealing with much more complex activity and give fair recognition
rate. However, HMM recognizer is not involve feature priority
that should has effect to accuracy so we proposed the third
recognizer that used graph similarity measurement for activity
modeling and activity classification. The third one, Graph
Similarity Measurement (GSM) recognizer involved feature
priority for recognition method then show better result than
HMM in most measurement. GSM recognizer has ~84% accuracy
in average. FSM recognizer is suitable for simple activity with low
computation cost while HMM is suitable for much more complex
activity and use single feature for recognition process. However,
HMM method may not give best result for the activity that use
multiple features. GSM is also suitable for complex activity and,
furthermore, give better result than HMM for the activity that
trained from multiple features.