Back to articles
Volume: 30 | Article ID: art00018
Raindrop detection considering extremal regions and salient features
  DOI :  10.2352/ISSN.2470-1173.2018.17.AVM-348  Published OnlineJanuary 2018

Images captured from vehicle mounted cameras endure various uncontrollable adverse conditions such as rain, dust, smudges etc. These result in artifacts which corrupt the captured scene globally or locally. Raindrop is a commonly observed occlusion in images captured from behind a windshield on a rainy day. While reconstruction is a challenging task, detection of these artefacts is also a non-trivial task as there is no well-defined singular model of raindrop or artifact created by it; neither there is a fixed shape, blur and glare level. In this work we address the detection of artefacts caused by raindrops. We employ a structural verification approach to identify the regions and exploit the uniformity that can be observed in the occlusion region caused due to collection of light rays in the drop by using maximally stable extremal regions (MSER) on the frame to get initial estimate of the regions. False positives are discarded by filtering the estimates based on area, orientation, eccentricity, roundness and convexity. From our results and observation we can conclude that raindrops indeed form extremal regions, however, detection accuracy using MSER is highly susceptible to false positives. Accuracy can be greatly improved using the proposed techniques on the regions based on our observations. A precisionrecall analysis is performed to assess the performance of the method.

Subject Areas :
Views 9
Downloads 0
 articleview.views 9
 articleview.downloads 0
  Cite this article 

C. S Vijay, Radhesh Bhat, Vijaya Ragavan, "Raindrop detection considering extremal regions and salient featuresin Proc. IS&T Int’l. Symp. on Electronic Imaging: Autonomous Vehicles and Machines,  2018,  pp 348-1 - 348-6,

 Copy citation
  Copyright statement 
Copyright © Society for Imaging Science and Technology 2018
Electronic Imaging
Society for Imaging Science and Technology