基于改进GMM算法的林火烟雾识别研究

针对森林火灾烟雾场景光照突变时传统混合高斯模型(GMM)无法适应的问题,提出基于改进GMM的林火烟雾识别算法。通过六足机器人平台上的CCD摄像机读取当前帧图像,与混合高斯模型建立的前一个背景图像作差分得到变化区域,计算二值化后的变化区域中像素值为1的像素点占总像素点的比例,与设定的阈值相比较从而判别森林火灾场景中的光照是否发生突变。若场景内光照发生突变,采用一个较大的更新速率α,保持模型的稳定性;若场景内光照未发生突变,根据不同的情况自适应调整更新速率α的值,保证模型快速收敛。实验表明,改进的GMM算法可检测到场景中相对完整的动态烟雾区域,满足林火烟雾的检测要求,以期为图像处理在森林火灾巡检机器人上的应用和进一步研究提供参考。

Forest fire smoke identification algorithm based on improved GMM

In order to solve the problem that the traditional mixed Gaussian model cannot adapt to the sudden changes in the light of the forest fire smoke scene, an improved fire smoke identification algorithm is proposed. The current frame is read by the camera based on the hexapod robot. The difference image between the current frame image and the background image created by the mixed Gaussian model is differentiated to obtain the change area. The ratio of the total pixel points occupied by the pixels with the pixel value 1 is calculated and compared with the threshold value. By comparison, the paper determines whether light mutations occur in forest fire scenes. When a sudden light change occurs suddenly, a large update rate is required to quickly restore the scene to stability. When the light does not change suddenly, the update rate is dynamically adjusted according to different situations to ensure rapid convergence of the model. Experiments show that the improved hybrid Gaussian model algorithm has achieved a good detection effect. It has strong adaptability and robustness in forest fire scenarios, which can meet the real-time detection of forest fire smoke and in order to provide reference for the application and further research of motion detection in forest fire inspection robots.