A low-light image scene understanding method

Abstract

The invention belongs to the technical field of computer vision, and provides a low-light image scene understanding method. A frame of series model of scene understanding based on depth learning and image enhancement is designed. For the input low-light image, the image is enhanced first to improve the brightness and discrimination without changing the color information of pixels in the image. Inthis framework, the task of scene understanding in low-lighting images is solved. The visual enhancement network provided by the invention can improve brightness on the basis of keeping picture colorinformation through jump connection operation, combining low-level color and structure information and high-level semantic information while restoring details, thereby restoring picture details. The restored image will effectively facilitate the detection, segmentation and other scene understanding tasks. Comparing qualitatively and quantitatively, the method proposed by the invention is superiorto the existing picture enhancement methods in both synthetic and real pictures.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a method for improving scene understanding under low light using an image enhancement technology based on deep learning. Background technique [0002] Scene understanding (Scenes Understanding), such as object detection and semantic segmentation, is playing an increasingly important role in the field of robot navigation and autonomous driving, because it can provide key visual information for behavioral decisions of robots or vehicles. However, in previous work, existing scene understanding methods mainly focus on clearly recognizable images or video frames captured in sunny weather, but for images captured in bad weather conditions or at night, due to insufficient illumination The loss of details, unclear edges, and pixel distortion of the captured image will lead to a serious reduction in the quality of image scene understanding. This kind of scene is often...

AI Technical Summary

Problems solved by technology

[0011] Aiming at the problem that the existing scene understanding method can only operate in an ideal environment, and it is difficult to achieve good results in a non-ideal environment such as low light, the present invention designs a series model of scene understanding based on deep learning technology combined with an image enhancement method The framework, for the input low-light pictures, can first enhance the picture, improve the brightness and recognition of the picture without changing the pixel color information in the picture, and then perform the task of scene understanding

Images