Zemūdens attēlu uzlabošana reāllaika video

Abstract

Ūdens videomateriāli spēlē nozīmīgu lomu jūras un okeānu pētniecībā. Zemūdens video dati bieži vien ir ar ļoti sliktu redzamību, kas apgrūtina efektivu pētniecību. Pēdējos gados ir izvirzīti dažādi veidi, kā uzlabot zemūdens video kvalitāti, izmantojot attēlu apstrādi, bet lielākajai daļai no tiem ir ierobežojumi – metodes labi strādā tikai konkrētos apstākļos, un attēla apstrādes laiks ir par lielu, lai to izmantotu reāllaikā. Šajā pētijumā tika pārskatitas daudzas esošās metodes ar mērķi ieviest metodi, kas ne tikai varētu uzlabot zemūdens video kvalitāti līdzīgi esošajām metodēm, bet arī to veikt reāllaikā. Veiktā aptauja liecina, ka 81.67% no individuāliem vērtējumiem novērtēja izveidotās metodes uzlaboto video attēlu kā labāku, salīdzinoši ar oriģinālo. Attiecībā uz veiktspēju katrs video kadrs tiek apstrādāts 0.18 sekundēs, un ar pareizu realizāciju tas var tikt izmantots reāllaikā. Bakalaura darbs rakstits angļu valodā. Tas sastāv no 48 lapām. Darbā izmantotas 34 atsauces, 12 ilustrācijas un viens pielikums. Atslēgas vārdi: Zemūdens attēlu uzlabošana, zemūdens video uzlabošana, reāllaika video uzlabošana.

Underwater video data plays a huge role in how we understand the marine life. Many of underwater exploratory methods, such as ROV operations, involve filming underwater videos that are used for further analysis, or they rely on real-time video data to perform various tasks; however, the video data filmed underwater is often with very poor visibility due to the nature of how light travels in underwater environment. For this reason, a lot of the information about the underwater world is gathered using other methods, such as acoustic sonars instead of traditional cameras. The downside is that these methods are very costly, and a lot of research and exploration is limited due to the costs. In recent years, methods to improve underwater video quality through image processing have been proposed, but most of them have limitations. The most common problem is that many of the methods work as expected in specific light or color conditions; therefore, they do not yield optimal results most of the time. Another limitation is the processing time of the algorithm. Even if some algorithms are capable to improve the quality of the video by a lot, they often are only applicable as a post-processing tool, which means that they cannot be used in real-time operations. For this research, many existing methods were reviewed with the aim to implement a method that can not only improve underwater video quality to the level of existing methods, but also do it in real time. Nearly 5 hours of diverse video data was collected. The proposed method consists of 3 parts – Dehazing, Color Correction and Contrast Enhancement, and the method contains two configuration parameters – white balance and clipLimit. This means that the method is flexible, and the configuration parameters allows to achieve the best results in variety of conditions. The method was evaluated with various metrics, and to validate the findings from the metrics, a survey was conducted. Survey results show that 81.67% of individual assessments preferred the enhanced video data, which indicates that the method is improving the underwater video quality. In terms of performance, each video frame is processed in 0.18 seconds, and with proper implementation, it can be applied to real-time video feeds. That will enhance and display the video with the rate of 5.5 frames per second. Not all image quality methods were tried; therefore, there is the potential for further research to increase the enhancing quality and the speed of enhancing and ultimately find more suitable methods for the challenges of real-time underwater videography. The paper is written in English. It consists of 48 pages. It contains 34 references, 12 figures, and one appendix. Keywords: Underwater image enhancing, underwater video enhancing, real-time video enhancing.