A Deep Learning Approach for Detection and Analysis of Anomalous Activities in Videos

The detection and analysis of anomalous activity in videos have become increasingly important in many real-time applications like patient monitoring, passenger monitoring, and public safety. Differentiating between normal and abnormal human activities poses a significant challenge due to their inherent similarities and challenges like occlusion, clutter etc. This proposal aims to develop a system which can identify and distinguish multiple anomalous actions from video sequences which has occluded scenarios, both individually and in groups. It effectively differentiates between various normal and abnormal activities in the given input videos. Toachieve, this proposal employs a combination of Convolution Neural Networks (CNN), Long Short-Term Memory (LSTM), and Long-term Recurrent Convolutional Networks (LRCN) models. Created own data sets for the required scenario along with existing large datasets UCF101, HMDB having patient and passenger videos are used to train the model. The experimental results demonstrate that the proposed approach achieves an accuracy of 96.8% in recognizing abnormal actions. This model is potential to use in the applications like hospitals, elder care homes, and public places where accurate monitoring of human activities is crucial and by efficiently detecting and identifying abnormal activities, it can aid in the early recognition of potential risks of unusual behavior, facilitating timely intervention and ensuring the safety and security of individuals in such environments.