ERDD: efficient rain removal algorithm in surveillance videos via rain detection and diffusion models

This paper introduces a novel rain removal algorithm aimed at improving the visual quality of images and videos affected by rain. The core of ERDD lies in its sophisticated rain detection mechanism, which meticulously identifies and isolates the impact of raindrops. Subsequently, a state-of-the-art diffusion model is applied to refine and optimize the imagery, effectively mitigating the rain's visual interference. Comprehensive evaluations conducted on a variety of benchmark datasets reveal that ERDD outperforms existing methodologies in rain removal, particularly excelling in the preservation of image details and color accuracy. The results from the rain detection conducted on 1800 rainfall images showed that the average processing time for each image was 0.014 seconds, with an accuracy of 96.61%.