Nayantara: Crime Analysis from CCTV Footage Using MobileNet-V2 and Transfer Learning

Traditional closed-circuit television (CCTV) systems are limited in their ability to detect and respond to criminal incidents without human intervention. Advanced CCTV analytics solutions can overcome these limitations by using machine learning and deep learning techniques to automatically analyse video footage and detect criminal incidents. These solutions can identify a variety of suspicious activities, track the movement of individuals and objects, and identify objects that are out of place. In this paper, we propose the design of a new-age lightweight efficient CCTV analytics system, Nayantara. The proposed system utilizes neural networks to identify suspicious criminal activities and track the movement of individuals and objects. It also includes an agent-based web application that allows security personnel to view the live video footage and receive alerts about potential criminal incidents detected by the model. The proposed model was trained and evaluated using the UCF-Crime dataset. The results showed that the solution was able to detect criminal incidents with 85.6% accuracy. In addition, Nayantara incorporates the provision for crime reports and weekly crime statistics generation to provide a holistic visualization of the crime landscape, thus, promoting better decision-making. Nayantara has the potential to significantly improve the ability of CCTV systems to detect and respond to criminal incidents. This could lead to a reduction in crime rates and an improvement in public safety.