Abstract

Deep learning has shown significant performance in many domains including natural language processing, recommendation systems, and self-driving cars in current years. From all the available applications detecting anomalies is a key problem that has been studied within research domains. The purpose is to assists with recognizing individual actions and detecting whether it is an anomaly or normal activity. To address this challenge of a detection algorithm for action recognition the author has presented a 3dimesional convolutional neural network model to detect real-time anomalies using drone surveillance. The initial stage of the model extract features from each person in the video and represents the data. Analysis of each extracted sequence to detect the associated actions is also proposed. Access dataset and UCF- Rooftop dataset was used for training and testing purposes. The results of this work revealed that the proposed 3dimensional convolutional neural network method provides more accurate anomaly detection achieving around 90%.