AI-Powered Smoking Detection System Using MaixCam and YOLOv5

The AI-Powered Smoking Detection System operates as an edge-computing vision platform using the MaixCam development board to run real-time inference without relying on cloud servers. The MaixCam board integrates a specialized neural network accelerator capable of executing a lightweight custom YOLOv5 model, which has been compiled specifically for edge devices (such as a smoking.cvimodel file). When the system boots up, it runs a custom Linux operating system environment executing Python 3.11 scripts to manage the camera feed and inference loop. The camera captures continuous high-resolution frames of the monitored environment. Each frame is passed directly into the quantized YOLOv5 object detection model loaded in memory. The model evaluates the visual data to identify patterns, shapes, and contexts characteristic of cigarettes and active smoking behavior. If the detection confidence exceeds a pre-configured threshold, the system triggers local hardware responses using its general-purpose input/output (GPIO) pins. Specifically, it can light up indicator LEDs to show system status or detections, and drive a buzzer to sound an immediate audible alarm. Additionally, the system captures snapshot images of the violation for evidentiary or logging purposes. Powered by a stable 5V power adaptor, the entire standalone setup continuously monitors enclosed areas like office cabins or public restrooms, processing video streams locally and reacting instantaneously to detected infractions.