Convolution Neural Network-based Action Recognition for Fall Event Detection
Mohd Fadzil Abu Hassan · International Journal of Advanced Trends in Computer Science and Engineering · 2019
Action recognition is a challenging and essential task in computer vision particularly in monitoring elderly activities in daily living.This paper presents an efficient normal and abnormal actions recognition model, particularly for fall event detection using the convolution neural network; CNN-AlexNet.In this study, the model was trained and tested using an established CASIA and Le2i datasets that cover various application domains.Results showed that the model performed well with classification accuracy, F-score, sensitivity and specificity rates of 99.97%, 99.97%, 99.94% and 100%, respectively.Therefore, the CNN-AlexNet based action recognition model can be considered as a high-performance classifier to differentiate between normal and abnormal actions to be adapted in a smart camera-based surveillance system.