AI-Based Smart Surveillance System Using Raspberry Pi and Embedded Vision

Maintaining real-time security in residential, commercial, and industrial environments, smart surveillance systems that are able to discern and identify objects and suspicious actions correctly are necessary. Conventional security systems that use human surveillance or basic motion sensors do not offer timely and reliable notifications. The paper offers an AI-based Smart Surveillance System that runs on the Raspberry Pi, embedded vision, and an Adaptive Object Recognition and Classification Model, which is based on YOLOv5 with Convolutional Neural Networks (CNNs) and multi-scale prediction layers. The model is dynamically designed to change the detection thresholds according to the changes in the environment in terms of lighting, background, and distance of the object, which allows it to accurately identify and categorize human beings, vehicles, animals, and other objects in real-time. The system records visual and spatial data with the help of a Camera Module, PIR motion sensor, and ultrasonic sensor and processes it with the help of DCSASS Dataset. The real-time notifications are provided by the use of a buzzer, LED indicators, and LCD display which guarantee instant knowledge of danger. The experimental findings demonstrate high reliability, low latency, and detection accuracy of 96.8 percent, and are also strong in the complex environments. Adaptive deep learning, embedded vision, and edge computing integration is a scalable, cost-effective and automated solution to modern smart surveillance applications.