Combustion processes in industrial furnaces pose challenging difficulties when one attempts to model flame. Chemical and physical phenomena interact and affect flame behavior in such a complex way that several different approaches have been under development to tackle the modeling problem. Flame instabilities could potentially harm safety operation of the furnace. Ultimate goal of flame control is to develop means to actuate over the flame state in order to avoid dangerous situations. A critical step towards this objective is to come up with a reliable model for the flame, so that control theory could be applied to automatically maintain flame state within safe standards. This work proposes a methodology to model the dynamics of flames through computer vision to acquire and process flame images in order to extract image features evolution in time. Then, a dynamic model is identified through random decrement algorithm and Ibrahim time domain method, which are operational modal analysis techniques, together with a random contribution. Flames from seven different combustion conditions were modeled by such methodology, and their estimated values were compared with experimental data for validation. Results show that estimated and experimental data possess a high degree of correlation, thus confirming the viability of the proposed dynamic model of flames.