Smart window system for air quality control and threat detection based on deep learning

Purpose This study aims to develop an integrated Smart Window System to enhance indoor safety and comfort by autonomously addressing two critical threats: fire hazards and poor indoor air quality (IAQ). Design/methodology/approach The proposed system combines an ESP32 microcontroller and a Raspberry Pi within an Internet of Things (IoT) architecture. The ESP32 manages real-time environmental monitoring using DHT11 (temperature, humidity) and MQ-7 (carbon monoxide) sensors and controls actuators (window motor, DC fan and buzzer). The Raspberry Pi runs a YOLOv9 deep learning model locally on a camera feed for real-time fire detection. A mobile application provides user oversight and manual control. The system’s performance was validated through experimental testing, measuring detection accuracy, response times and IAQ mitigation effectiveness. Findings The system achieved 94% mean Average Precision ([email protected]) in fire detection, with an end-to-end response time of 4–6 s from fire detection to window actuation and alarm triggering. In IAQ management, the system successfully maintained parameters within safe thresholds, reducing elevated CO levels by 40% within 10 min by activating ventilation. The mobile interface demonstrated reliable control with less than 1-second latency. Research limitations/implications The main limitations concern the use of a single benchmark data set and the restricted duration of environmental testing. Future work will expand real-world validation across diverse lighting and smoke conditions, integrate higher-precision sensors (e.g. PM2.5/VOC modules) and enhance resilience through GSM-based alerts and solar-assisted power management. Practical implications The proposed system provides a low-cost, fully automated framework for smart buildings, enhancing fire safety and IAQ without reliance on cloud infrastructure. Its modular hardware–software design enables seamless integration with existing electrical and ventilation systems, supporting scalable deployment across residential and commercial environments. Originality/value This work presents a novel, fully integrated solution that seamlessly merges deep learning-based computer vision for high-accuracy fire detection with IoT-based environmental sensing and actuation. Unlike previous systems that address these problems in isolation, this design enables a unified, autonomous and real-time physical response to both immediate fire threats and gradual air quality deterioration within a single, low-cost framework.