Suspicious Human Activity Detection Through Video Analysis and Alert System: A Comprehensive Review

Human activity detection, a vital computational method for video analysis, aims to discern ongoing events within video material. Specifically, this review paper focuses on the process of identifying inappropriate human activities—often categorized as suspicious actions—such as leaping, climbing, and crawling. Leaping, climbing, and crawling in restricted or restricted areas might signify security breaches, accidents, or emergencies. This analysis involves the segmentation of live video into frames to scrutinize human behavior. Detecting humans within video data has long been challenging due to the non-rigid, unpredictable nature of the human body, exacerbated by factors like poor lighting and varied stances. Timely alert generation systems are imperative for notifying system users when suspicious activity is identified. The review delves into state-of-the-art models, with You Only Look Once (YOLO) emerging as a popular choice. YOLO offers a realistic and versatile framework with multiple versions suitable for diverse applications. These models process video input by splitting it into frames and conducting object detection based on their training data. Additionally, transfer learning is explored as a method to enhance model performance by training on custom datasets, facilitating real-time detection of normal and suspicious human activities. Lastly, the review discusses practical means of alerting system users, with a focus on the efficiency of Gmail notifications in promptly notifying relevant personnel about potential security or safety concerns.