Algorithm for forest fire smoke detection in video

• Main Authors: Rykhard P. Bohush, Sergey V. Ablameyko
• Format: Article
• Language: Belarusian
• Published: Belarusian State University 2021-04-01
• Series: Журнал Белорусского государственного университета: Математика, информатика
• Subjects: forest fire; image analysis; background; covariance descriptors; support vector machine
• Online Access: https://journals.bsu.by/index.php/mathematics/article/view/3652

In this paper, an efficient forest smoke detection algorithm in video sequences obtained from a stationary camera is proposed. The algorithm composed of three basic steps. At the first step, the frame contrast is improved. After that detection of slowly moving areas is performed based on dynamic and static features. For this we use adaptive background subtraction and color segmentation. The detected areas are divided into small blocks. Spatio-temporal analysis is applied to them. Blocks are classified based on covariance descriptors and support vector machine with a radial basis kernel function. Experimental results for processing real video show effectiveness of our algorithm for early forest smoke detection.