Information Inversion and Dynamic Analysis of Video-Driven Fire Detection Based on Object-Oriented Segmentation

During an emergency response to a sudden fire, the video-driven techniques for detection and real-time analysis can significantly affect the scientificity and effectiveness of disposal, thus changing the process of fire rescue. In this paper, an information inversion model for video-driven fire based on object-oriented segmentation has been proposed and programmed with Python. Firstly, a video fire detection algorithm based on deep learning was adopted. Secondly, fire regions were segmented using the SEEDS superpixel-based method, from which the flame size with time was further extracted. Then the heat release rate was inversed over time. Finally, the fire development was analyzed based on the characteristics of flame images. The model was applied to three different types of fire, a diesel pool fire, a propane fire, and a full-scale wood crib fire. The results indicated that the model could achieve real-time information such as flame height, projection area, and volume, and the spatial–temporal evolution laws could be further estimated quickly. Committed to intelligent analysis of fire environment based on surveillance cameras, the model can provide technical assistance for fire-fighting and on-site emergency rescue.