Using Artificial Intelligence for Anomaly Detection Using Security Cameras

Identifying suspicious activity in public areas is a common concern in machine learning studies. However, this task is not usually trivial. In this article, we propose a method for recognizing anomalies in the public transportation environment to guarantee the safety of its passengers. The method in question consists of a classifier based on Convolutional Neural Networks (CNN). As input to this model, we use images from Closed-Circuit Television (CCTV) cameras already present in vehicles for security reasons. The addressed problem became challenging for reasons such as the lack of standardization of equipment, the low quality of the images provided, and the poor positioning of the cameras. In addition, a dataset, which has a high imbalance between the classes, was built. We evaluated four CNN architectures on the dataset to validate the proposed method. Experiments on the created dataset showed that the proposal of this project achieved promising results.