**6. Conclusions**

Traditional haze removal methods fail to restore the images with different degrees of haziness in a real-time and adaptive manner under most circumstances. To solve this problem, we propose an efficient traffic video dehazing method using adaptive dark channel prior and spatial-temporal correlations. The dark channel prior is based on the statistics of outdoor haze-free images, but it cannot adaptively estimate the initial transmission value based on the degree of haze and contrast of images. Therefore, we adopt the image-contrast-enhanced method to obtain the best estimated transmission value as the initial transmission value of dark channel prior. The image dehazing method using adaptive dark channel prior can overcome the shortcomings of existing dehazing algorithms that overstretch contrast after haze removal and deal with images with dense haze to a satisfactory level. Additionally, we introduce the temporal-spatial correlation of traffic videos to speed up the traffic video dehazing using the time continuity to set a time slice, the characteristics of block structure to refine transmission, lane space structure to decrease the restored area, and multi-camera distribution to simplify the calculation of parameters. The experiment results show that our method can restore satisfactory image appearance, which can remove dense haze effectively and does not produce results with overstretched contrast. The temporal and spatial characteristics can reduce the computation time, especially for dehazing multiple videos.

However, the dark channel prior is a kind of statistic, and it may not work for some particular traffic videos. When there are rapidly changing hazes in the videos, the dark channel of the scene radiance has a grea<sup>t</sup> difference at different times. In addition, if the scene objects are inherently similar to the atmospheric light and no shadow is cast on them, the adaptive dark channel prior is invalid. The dark channel of the scene radiance has bright values near such objects. As a result, our method may underestimate the transmission of these objects and overestimate the haze layer.

**Author Contributions:** Formal analysis, G.Z. and J.W.; methodology, T.D., Y.Y. and Y.S.; project administration, T.D.; validation, Y.Y.; literature search, J.W. and G.Z.; writing-original draft, T.D. and J.W.; writing-review and editing, G.Z. and Y.S.

**Funding:** This work is supported by National Natural Science Foundation of China (No. 61672414, 61572437).

**Conflicts of Interest:** The authors declare no conflict of interest.
