Wildfire smoke detection based on co-occurrence matrix and dynamic feature

Abstract

The paper presents a new approach to detect wildfire smoke in forest by using camera surveillance. The basic idea of the proposed method is that smoke is grayish and nonrigid object, it normally creates chaos information in image. The new approach includes three steps, the pre-processing step to reduce noise as well as to divide the image into small blocks, the determining candidate smoke objects step using smoke color detection and slow motion detection, and the analysing smoke objects step using Co-occurrence matrix to extract data from candidate smoke objects. The experiment results on wildfire smoke videos show that Co-occurrence matrix can be used to extract smoke features and it can be combine with other extracting features method to improve the accuracy of the wildfire smoke detection method.