Reflection Removal On The Windshield During Nighttime Driving

In recent years,with the continuous increase of car ownership,the risk of traffic accidents is also rising.And the automotive electronic products which are used to assist safe driving ushered in a broad development prospect.The driving recorder can record the surrounding environment of the vehicle and save various data of the car,which effectively protects the safety of the driver's personal property and is favored by consumers.It is favored by consumers.However,the quality of the on-board video is often degraded by the reflection layer appearing on the windshield.We need to remove the reflection layer to improve the readability of the car video.During the daytime driving,the reflection layer on the windshield is mostly the reflected image of the objects placed in the car.At present,corresponding algorithms have been developed to remove it from the video.However,the illumination at night is special,and the brightness in the car is lower than that outside the car.During the nighttime driving,the reflection layer on the windshield is no longer the reflection of general objects,but the external light transmitted and refracted by the glass and the dust on it.At present,no published papers have been found to solve this problem.We focus on this problem and attempt to improve the readability of video by removing the undesired reflection layer from the nighttime on-board video.We firstly apply gradient prior and the characteristics of reflection to automatically detect the reflection regions,the reflections in these regions are then classified into the major and minor reflection.Next,the Laplacian fidelity term and the L₀gradient are used to construct the objective function,and the function is optimized by semi-quadratic regularization and Adam optimization method.This allows the algorithm to reduce the gradient intensity of the main reflection area while maintaining important details in the image to achieve the purpose of suppressing the main reflection.Based on the different distribution of gradient histograms between the background layer and the reflection layer,the objective function is constructed,and the reflection removal problem is modeled to remove the main reflection residue and the secondary reflection.Finally,the resulting reflection removal result is added to the non-reflection region to obtain a reflection removal result.The algorithm is tested by performing experiments on images are which from night-time on-board video and the daytime SIR²standard data set.And subjectively and objectively compared and analyzed with the state-of-the-art.The results show that our proposed algorithm performs better in reflection removal than others.Besides,its performance of keeping details,color,and brightness of the original image is satisfied and faster than most existing algorithms.