Underground target detection algorithm based on improved Gaussian mixture model

Abstract

The monitoring video images of underground coal mine have problems such as poor quality, noisy and being susceptible to sudden changes in illumination. The traditional Gaussian mixture model for target detection has problems such as slow running speed, high algorithm complexity and susceptibility to illumination. In order to solve the above problems, an underground target detection algorithm based on improved Gaussian mixture model is proposed. The improved dark channel defogging algorithm is applied to preprocess the underground image, finding the dark channel map for the thumbnail of the underground fog map, and using bilinear interpolation to obtain the defogging image. Based on the Gaussian mixture model, an improved block modeling strategy is used to reduce the modeling complexity and improve the algorithm running speed. Combined with the three-frame difference method, different learning rates are set for the early and late Gaussian modeling according to the proportion of the image foreground to suppress the influence of illumination on target detection and improve the modeling speed and accuracy. The experimental results show that when the illumination changes suddenly, the algorithm proposed in this paper can still describe the detection object well, and has a significant suppression effect on illumination changes. Compared with the three-frame difference method and the traditional Gaussian mixture model, the proposed algorithm can improve the processing speed effectively.