Efficient ATM-based Anomaly and Suspicious Activities Detection with Automated Alert using Enhanced Super-Resolution GAN and RoI based Detection

The Automated Teller Machine (ATM) is a vital source for daily monetary transactions. The public uses ATMs, which should be safe and secure for people to use. However, there may be scenarios when some anomaly activities happen in ATMs, such as threats and wearing helmets, involving abnormal activities. This work aims to develop an automated ATM security system using YOLOv10 for real-time detection and classification of quickly identifying potential threats, such as fraudulent activities or individuals trying to conceal their identity, and triggering alerts for appropriate security responses. This methodology utilizes ESRGAN to upscale low-resolution images, improving the clarity and detail in security footage. Preprocessing with ESRGAN, aspect-ratio preservation, and defining region of interest prepare the images for effective YOLOv10 training and detection. This approach ensures that imperative details are preserved for accurate detection while focusing on critical areas in the ATM environment, enabling a reliable anomaly detection system with efficient triggering and minimal false alerts. The proposed system detects the anomaly and suspicious activities in the ATM environment with an accuracy of 99.05%. Thus, the work is significant for detecting thefts by people in real-time trying to hide their faces with helmets and other suspicious activities.