Smoke Recognition Based on Machine Vision

In fire protection field smoke recognition is becoming more and more significant compared with traditional method. Vision-based smoke detection approach using image processing technology in smart monitoring system is proposed for early alarm. Firstly, to improve the image quality, fuzzy logic is employed for image enhancement. Then candidate Smoke area is extracted by Gaussian Mixed Model. According to the features of smoke, static and dynamic features are extracted. Thirdly, hyper parameter of Support Vector Machine(SVM) is determined and then the model of SVM is established. Finally, the characteristics vectors are input into SVM for smoke recognition online. Experiments are made by several videos under a variety of conditions and it is shown that this proposed algorithm is feasible and effective. It also has strong robustness and high real time.