Deep Maxout Network for human action and abnormality detection using Chronological Poor and Rich Optimization

Abstract

ABSTRACT Human activity recognition targets to discriminate the various human actions depending on the size, shape and posture exhibited by the humans while performing a particular action. Video-based human action detection easily understands the actions and behaviours of humans in the video sequences. Therefore, it has a part in various applications, like human-machine interaction and healthcare domains. However, human activity recognition in a complex environment leads to poor performance and low accuracy, and it remains a challenging task in human action recognition using video sequences. To overcome such limitations, a Chronological Poor and Rich Optimisation (CPRO)-based Deep Maxout Network is developed for effective human movement and abnormality recognition. The CPRO is devised by integrating the chronological concept into Poor and Rich Optimisation. In the feature extraction process, the features include Histogram of Oriented Gradients, Local Gradient Pattern and hierarchical skeleton are digging out. Moreover, human action and abnormality detection are done utilising Deep Maxout Network trained by the proposed CPRO. The proposed scheme reveals superior results with respect to evaluation measures, such as accuracy, sensitivity and specificity of 0.954, 0.958 and 0.960, respectively.