Target-Tracking Based Early Fire Smoke Detection in Video

Abstract

The paper proposed a target-tracking based fire smoke detection method for early fire-alarming system at large or open space. The method utilizes an improved Gaussian mixture model positioning algorithm, an efficient target tracking algorithm as well as three effective static and dynamic smoke visual features: brightness consistency, motion accumulation and spread. Finally, an algorithm combined temporal and spatial information to assess the fire alarm algorithm is implemented by considering the performance requirements of the fire-alarming system. Experimental results show that the method has low time complexity and is able to rule out the major interference sources.