Silhouette-Based Haman Action Recognition Using Convolutional Neural Network

Nowadays BAR is gaining interest in computer vision area. The BAR system was utilized and applied in various applications related to surveillance. Some of the surveillance applications are airport surveillance, shop surveillance, bank, schools, and many areas. The goal is to recognize human actions automatically. If the raw video is taken as input without any preprocessing then the human action recognition system consumes more time and leads to increased storage space. So in this work silhouette is prefered to extract the features with a time and space consuming manner. and finally CNN is used to classify the actions. The standard Weizmann dataset and KTH dataset is used to evaluate and manipulate the performance of the proposed HAR technique. The proposed method provides 98 percent accuracy for the Weizmann dataset and 93 percent accuracy for the KTH dataset. The HAR technique provides efficient outcome than the recent existing techniques that are used for used for human action recognition(HAR) systems.