This technology uses a deep neural network to restore images captured in the dark. By preprocessing images acquired under low and non-uniform lighting conditions, we can improve the performance of object detection substantially. Experimental results on the VisDrone2019 dataset demonstrate the effectiveness of the proposed method, achieving a remarkable 5% increase in average recall.

Despite abundant past research on low-light image restoration, little emphasis has been placed on how the restored images enhance the performance of object recognition. The proposed image restoration mechanism can effectively improve the image quality taken in the low light condition, and can also be generalized to handle other types of degradation, including over-exposure, multiple light sources.

The proposed framework improves the quality of acquired images, as well as the accuracy of the subsequent object detection process. The developed technology is applicable to all tasks that require automatic analysis of image content, including video surveillance, navigation of unmanned autonomous vehicles, and search/rescue missions.