Research on Video Smoke Recognition Based on Dynamic Image Segmentation Detection Technology

Abstract

The smoke feature recognition stage is a relatively core part. The quality of the smoke feature recognition algorithm directly affects the accuracy and real-time of the alarm. The existing smoke feature recognition methods are roughly classified into the recognition of color features, the identification of static features of smoke, and the identification of dynamic features of smoke. Experiments show that using the cumulative amount and the main motion direction to identify the combined motion history image and the constant distance to identify the characteristics of the smoke can get good results. Digital image processing technology and pattern recognition technology enable the system to sense where the human eye can perceive and detect fires based on the characteristics of the smoke image. In practical applications, video smoke detection only requires hardware devices such as cameras, and even the use of its own security surveillance camera, the hardware cost is very low; and the scope of use is very wide, basically no conditions are limited; portability is very good. This paper combines fire detection and recognition based on video images with visual attention mechanisms. Preprocessing the video image using the visual attention mechanism not only compensates for the shortcomings of the traditional fire detection method, but also significantly improves the speed and accuracy of fire detection and recognition.