Suspicious Activity Detection System Using Machine Learning

Monitoring human activities is an important part of ensuring public safety. The aim is to detect suspicious behavior. To achieve this, we use LRCN, a long-term continuous regression algorithm, to identify negative functions. When analyzing suspicious behavior, it is important to consider the temporal data of these videos, using a blend of CNN and RNN to explore the frames and extract relevant features. The main activities in the project include research, collecting and processing preliminary data, creating and training models, and evaluating their performance. The detection result can be accurate and detect suspicious behavior immediately. We use the KTH dataset containing 600 walking and running frames and the Kaggle dataset containing 100 training images to build the model. Our test models show that the system and video can detect suspicious situations with 86% accuracy, and we expect this accuracy to increase as the data grows.