Abstract: The video surveillance system is widely used in the comprehensive mining work of coal mines, and the quality of the images captured by video surveillance directly determines the safety and production efficiency of workers. Currently, it is affected by the uneven distribution of dust, water vapor, and light sources in the underground, and the collected image information is prone to problems such as large fog, overexposure, and low contrast, the traditional dark channel prior defogging algorithm has the advantages of fast calculation speed and good repair effect, so it is widely used in underground fog. However, when processing images with large dust and uneven distribution of light sources in coal mines, it can lead to a large amount of image information loss and damage to the texture of the light source area. To address this issue, an improved downhole fog algorithm for estimating the transmittance distribution is proposed. By utilizing a dark point light source model, combining the credibility weight factor of the dark channel with the optimal dark channel defogging image, the transmittance distribution is improved and the underground foggy image is defogged. The experimental results show that the peak signal to noise ratio, information entropy, standard deviation and average gradient of the improved transmission distribution estimation defogging algorithm are 16.25%, 31.39%, 17.17% and 48.47% higher than those of Retinex algorithm, respectively, and 47.24%, 35.24%, 21.64% and 51.79% higher than those of the traditional dark channel prior defogging algorithm. The experimental data shows that the improved transmittance distribution estimation and defogging algorithm reduces the loss of texture details in underground images, and effectively improves the contrast and clarity.