A Zero-DCE-Based Low-Light Image Enhancement Algorithm for Urban Roads

The quality of monitoring data sources has an important impact on road recognition and human vehicle detection performance in the field of smart transportation. Aiming at the problem of large light disturbance and low contrast in urban road monitoring images in practical applications, this paper proposed a method for low-light image enhancement of urban roads based on Zero-DCE. First, for the constructed urban road low-light image data set, an improved low-light enhancement network model was designed based on the idea of no-reference depth curve estimation; then, the model hyperparameters are ablated, and end-to-end training can be achieved without relying on paired training data, which overcomes the shortcomings of existing image enhancement algorithms that rely too much on high and low quality paired images in the same scene. The experimental results show that the method proposed in this paper can significantly enhance the brightness and contrast of the image, improve the visibility of details, and improve its subjective visual quality. At the same time, in terms of objective evaluation indicators and running time, this method also has certain advantages compared with traditional methods and other learning-based methods.