Smoke Detection for Videos Based on Adaptive Learning Rate and Linear Fitting Algorithm

In the process of smoke detection, the brightness and texture are varying in each frame, which brings interference to the feature extraction and reduces the accuracy of smoke detection. In this paper, the adaptive learning rate model associated with those noise objects is established, which provides a good regularization of background adaptation for Gaussian Mixture Models (GMM). Next, the linear fitting method is combined with the smoke motion and shape features to detect smoke objects. Experiments have shown that the algorithm can adapt to the illuminative changes in various scenes and detect the smoke objects with a very low false alarm rate.