Malicious Human Activity Detection

Malicious Human Activity Detection (HAD) is crucial for applications such as security monitoring, abnormal behavior detection, and intelligent surveillance systems. This study leverages high-tech wearable devices to develop and test a machine learning model that can accurately detect malicious activities, such as indulging in fights and unauthorized use of mobile phones on college premises. Deep Learning, a specialized branch of Machine Learning utilizing Complex Artificial Neural Networks (ANN s), is frequently employed to achieve high accuracy in identifying human activities from smartphone data. In this paper, we propose a model that combines Convolutional Long Short-Term Memory (ConvLSTM) networks and Long-term Recurrent Convolutional Networks (LRCN) to detect human activities. ConvLSTM networks are adept at handling spatiotemporal data, making them suitable for video-based activity recognition, while LRCNs integrate convolutional layers for spatial feature extraction with recurrent layers for temporal sequence learning. This hybrid approach allows the model to capture both the spatial and temporal dynamics of activities more effectively. The model is trained and validated using a substantial collection of videos from the publicly available UCF50 and HMDB51 datasets. These datasets are renowned for their extensive variety of activity attributes, encompassing diverse human actions captured in different contexts and environments. This robust statistical model showcases the potential of combining wearable technology and advanced machine learning techniques to bolster security measures and improve abnormal behavior detection in various applications.