In today's world, video surveillance is critical. Machine learning, artificial intelligence, and deep learning were all integrated into the system at the same time, and the developments accelerated dramatically. Different systems use the above variants to help distinguish between various suspicious activities and live video tracking. Human behavior is the most uncertain, making it impossible to decide whether it is suspicious. A deep learning approach based on face region recognition is used to detect the behavior of humans is normal or suspicious, and if suspicious activity is expected, an alert message is sent to the proper authorities in an application of ATM room. The extraction of consecutive frames from a video is a popular method of monitoring. The architecture is divided into two groups. Video frames are used to compute the features in the first section, and the classifier uses the features to predict whether the class is suspicious or normal in the second part. The structure of the face region is examined in order to determine whether the activity is suspicious or not.

—Suspicious activity, Head detection, Haar Features, Convolutional Neural Network (CNN), Video Surveillance