Retinex-Based Visibility Enhancement System for Inclement Weather with Tracking and Distance Estimation Capabilities

Road conditions affected by weather are well known to have an impact on the number of vehicle accidents and fatalities, due to low- to no-visibility conditions. According to the U.S. Department of Transportation, there are more than 1,259,000 crashes each year. On average, 6,000 people are killed and more than 445,000 people are injured annually due to severe weather conditions. These accidents could be significantly reduced if real-time visibility enhancement systems were made available. However, eliminating the impact of weather conditions on visibility is still lacking and beyond our control. The time has come to develop technology that is capable of improving visibility and creating safe driving conditions. It is the goal of this study to improve visibility and enhance drivers’ safety during severe weather and poor visibility conditions. When capturing images of inclement weather conditions, the light that reaches the camera is severely scattered by atmospheric obstacles (e.g. fog, rain, and snow), resulting in the degradation of the contrast quality. Depending on the nature of the distortion, or the environment conditions, the system can be custom-designed to enhance the visibility of the captured images and improve safety. Moreover, going through poor visibility can be seriously dangerous, since drivers may lose perception of distances, objects’ orientations relative to a focal point, and/or the depth of objects. In this study, Retinex technique was selected as the basis framework for developing a system capable of enhancing visibility for drivers. This technique was used due to its ability to achieve a good dynamic range compression and spectral rendition. These unique features, when properly deployed in a framework, can overcome the loss of background details. An innovative system is proposed through a multistage framework that not only incorporates a modified Retinex technique, but also uses object detection and depth estimation to overcome some of the current algorithms’ and systems’ drawbacks. The performance of the proposed system, along with histogram equalization and the basic Retinex enhancement techniques, are presented. Performance was assessed using Peak Signal to Noise Ratio (PSNR) and Structural SIMilarity (SSIM) parameters. The results show that the proposed system outperforms the comparable methods and indicates the efficacy of the system under a variety of visibility degradations