Abstract
Human action recognition in computer vision has become dexterous in detecting the abnormal activities to fortify safe events. This paper presents an efficient action recognition algorithm which is based on local binary pattern (LBP). The implementation of this approach can be used for action recognition in small premises such as ATM rooms by focusing on the LBP feature extraction via spatiotemporal relations. We also focus on decreasing the descriptor values by normalizing computed histogram bins. The results through ATM dataset demonstrate the enhancement in action recognition problem under different extensions. The normalized features obtained are classified using random forest classifier. In our study, it is shown that normalized version of LBP surpasses the conventional LBP descriptor with an average accuracy of 83%.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have