Implementing real-time wildfire detection using lightweight object-detection models and machine vision sensor on Raspberry Pi 5: Fireframe, a practical framework

Abstract. The research field of small, lightweight object-detection models that are capable of real-time monitoring, particularly for the detection of wildfires, is highly popular. However, a quick overview of the literature reveals that while most research suggests lightweight models, it does not report results from tests conducted on platforms with limited computational power or frameworks that might enable practical applicability of the techniques. This leaves the algorithms without real-time usability tests. This study addresses the research gap, aiming to provide a robust and low-cost framework (Fireframe) for edge device deployment for wildfire smoke detection. Fireframe combines hardware (Raspberry 5 computer and Basler machine vision camera) and a trained object detection model with tasks that are performed in an operational loop (main thread). It can simultaneously record a live stream, analyze whether wildfire smoke is present, and display the findings. Fireframe is tested using two lightweight models (YOLOv10 and MobileNetV3), and the findings confirm its suitability for simulations and real-life action.