A Hybrid Approach for Detection and Removal of Raindrops Using k-means Clustering and Hough Transformation

Author Name: S P Kullarkar, S V Jain
Author Email: kullarkarsp@rknec.edu


The vision system hired various tasks such as steering; information gathering can be unfavorably exaggerated by ruthless climate conditions such as rain, haze. Rainy weather condition raindrops twig to a frame glass can substantially abolish the scene visibility. Spotting and eliminating rain drops assist many computer visualization bids. To address this problem we introduced hybrid approach for raindrops detection and removal from video. In our approach we use K-means clustering, canny edge detector, Hough transformation and hybrid Otsu plus median filter for the effective reclamation of Rain drops from the video. We use Otsu algorithm at the detection phase and the median filter at the removal phase for fast retrieval of the raindrop from the image. Segmentation of image is perform by k-means clustering results in highest correct clustering rate. In detection process, we implement the Otsu algorithm and Hough transformation algorithm to know the exact location of the raindrop and edge detection is done by using canny edge detector. The detected region is filtered by using median filter method. The estimation of performance is done by using the parameters like accuracy and time.


Canny Edge Detector; Image Segmentation; K-means Clustering; Median Filter; Otsu Algorithm;  Raindrops  Detection


In rainy seasons raindrops probably stick to screens, protecting guards and camera lenses. Navigation, data collection and surveillance used outdoor vision system and opposed by bad climate situations such as flood, haze and snow. These sticky raindrops block and mislead some areas of the image and performances of many applications such as tracing, feature detection, audio correspondence that are tainted. A vision system where top-mounted vehicle sensor with no wipers is available and where a hand-held camera is used in such type of system this problem is occurred. Raindrops twig to a windscreen degrade the visibility of a scene. To address this problem, algorithm has been proposed which are k-means clustering, Hough transformation and OTSU to detect and remove the raindrops so that the system can be preprocessed and performance is enhanced. Otsu and median filter algorithm are applied for the fast retrieval of Rain droplets from the single image. The objective of our research is to propose a system which detect and remove the raindrops using K-means clustering and Hough transformation. For that we are planning to implement K-means and hybrid Otsu plus median filter noise removal algorithm, so that image can be pre-processed and the performance is enhanced.

In research field many papers analyze the presence of raindrops from their local spatial-temporal derivatives. Identifying raindrops from image is very challenging because the raindrops have different shape. We are using the Hough circle transformation to know the exact location of the object and the OTSU algorithm is also help to find the position of it.

Unlike rainproof Objects, adherent raindrops are transparent, their appearance and thus depending on the environment the intensity values varies. They feel pain from blur to the camera due to their propinquity. Most raindrops generate dirty look. The paper consists of the detail information about raindrops detection and removal techniques, these are organized as follows. The proposed methodology of effective and accurate segmentation technique of rain drop recognition and elimination is described using K-Means clustering, Hough transformation, Otsu and median filter. Finally we analyze the result which depends on the best performing parameters and conclude.


The developed algorithm is based on Otsu and median filter gives better result than existing approach. The k-means clustering algorithm results in highest correct clustering rate. We implement the Otsu and Hough transformation to identify the raindrop in the image. A detected region filters by using median filter. The performance is estimated by using the parameters like accuracy, sensitivity, time.

54 total views, 1 views today

Download File

About the author: admin