YOLOv3-based human detection and heuristically modified-LSTM for abnormal human activities detection in ATM machine.

In existing works, accurately analyzing human activities is a complicated problem in public places. Consequently, the detection of human activities becomes challenging the computer vision technology. The major scope of the research is to develop an abnormal Human Activity Recognition (HAR) model using deep structured architectures for detecting the suspicious activities of humans in the ATM using the video surveillance system. The classification phase utilizes the enhanced deep learning approach named improved Long Short-Term Memory (LSTM) by optimizing certain parameters in LSTM by hybrid optimization algorithm for accurately classifying the normal and abnormal activities of humans. This hybrid optimization algorithm is developed and termed Hybrid Spider Monkey-Chicken Swarm Optimization (HSM-CSO) for achieving the effective performance of the deep learning-based classification. Hence, the designed HAR model in ATM is proven that it helps to improve the system performance and also give relief from prohibited activities or crimes and false alarms for humans. [ABSTRACT FROM AUTHOR]