3D Location of Indoor Fire Ignition with a Multilevel Constraint Based on Monocular Vision

Abstract

Accurate fire ignition spatial location methods can serve the automatic fire suppression based on video. Although current fire detection systems based on monocular surveillance videos can quickly detect fires, it is impossible to obtain the 3D position due to the polysemy of 2D images. To further promote the universal application of automatic fire suppression, we propose a 3D indoor fire ignition location method based on monocular vision. This is the first study on spatial orientation of fire based on monocular vision. First, the indoor scene is quickly reconstructed as the basis of the scene. Second, based on our previous research on fire detection, the refined position of the fire in 2D images and its topological relationship with the space object are analyzed. The hierarchical constraints from 2 to 3D are established for the spatial orientation of the indoor fire ignition point. The experimental results show that the average absolute error is only 4.82 cm and that the average relative error is 1.71%. In addition, our method can be embedded into the existing fire prevention and control system at a low cost, further promoting the development of intelligent fire prevention and control.