Self-Attention-Based Progressive Generative Adversarial Network with Orthogonal Convolutional Neural Network in ATM
P. Sivaprakash, Francis H. Shajin, P. Rajesh, Pushpa Mamoria · IETE Journal of Research · 2025
Real-time human activity monitoring is now essential for the security and surveillance of public buildings and bank automated teller machines (ATMs) due to the daily rise in criminal behavior. To solve the issue of online identification of abnormal activity in ATMs, a Self-attention-based Progressive Generative Adversarial Network fostered Object detection with Orthogonal Convolutional Neural Network based Suspicious Human Activities recognition in ATM (SAPGAN-OCNN-SHAR-ATM) is proposed in this manuscript. Initially, the input video is taken from the RGB-D-ATM-Dataset. The input dataset is given to the Improved Adaptive Morphological Filter (IAMF) for locating the motion region or region of interest (ROI). Motion detection is an important phase in wide-scene analysis for foreground and background detection. The self-attention-based progressive generative adversarial network (SAPGAN) is used for determining whether an object is human or not and receives these extracted features to detect objects. Then, the proposed AMGNN-OD-BSNN-SHAR-ATM model is implemented, and the performance of the proposed method is analyzed with various evaluation metrics, like Accuracy, Sensitivity, Specificity, F-Score, precision and computation time.