Virtual environment for synthetic smoke clouds generation

摘要

In this paper we propose a virtual environment for the creation of synthetic wildfire smoke frame sequences, able to simulate a distant smoke plume and to integrate it with an existing frame sequence. This work provides a virtual tool to measure the accuracy of existing image-based wildfire smoke detection systems without the need to produce real smoke and fires in the environments. The proposed algorithm uses a cellular model driven by the rules of propagation and collision to simulate the basic physical principles of advection, diffusion, buoyancy, and the response to external forces (such as the wind). Adverse environmental conditions like fog and low-light are also simulated, together with the introduction of noise in order to reproduce acquisition defects. The resulting frame sequences are then evaluated by using a smoke detection system, which shows that our method for virtual smoke simulation gives results comparable to real situations. The extracted data can then be used to increase the performance of smoke detection systems when few real data are available.