Research on Algorithm of Vision-based Night Visibility Estimation

摘要

Accurately achieve the luminance of light sources at night images is significantly important to vision-based night visibility estimation. In this paper, we propose a practical night visibility algorithm. This algorithm contains two main parts, light sources recognition and edge extraction. Firstly, we give the model of vision-based visibility estimation, and propose the framework of algorithm. Secondly, we give the method to extract light sources from multiple potential targets by the prior knowledge of space information. Then, we analysis the features of light source image, and explain the expanded temple to cut sub-image of light source, and induce in PS level set to segment the edge. Experiments show that, the light source average recognition precision is approach to 0.95 at the condition of moderate breeze, and compared with the manual segment, the precision of light source segment is approach to 0.99 at the condition of real visibility larger than 500m.