In MartinRoth/fogDec: Fog Classification

knitr::opts_chunk$set(
  fig.width  = 6,
  fig.height = 4
)

Example how to use visDec

Load required libaries

library(data.table)
library(visDec)
library(ggplot2)
library(doParallel)
registerDoParallel(cores=3)
library(imager)
library(changepoint) # functionality should be included in imager
library(maptools)
library(lubridate)

Computing basis image statistics

#filenames <- c(
#  list.files("~/Dropbox/SWI_FogDetection/OctoberDay", 
#             pattern=glob2rx("Meetterrein_*.jpg"), full.names=TRUE),
#  
#  list.files("~/Dropbox/SWI_FogDetection/NovemberDay",
#             pattern=glob2rx("Meetterrein_*.jpg"), full.names=TRUE))
filenames <- list.files("/net/bhw420/nobackup/users/haijde/DATA/AXIS214/Meetterrein/201512",
                        pattern=glob2rx("*_2015*.jpg"),
                        full.names=TRUE, recursive = TRUE)
detect.edges <- function(im,sigma=1) {
  # adapted from http://dahtah.github.io/imager/foreground_background.html
  isoblur(im,sigma) %>% imgradient("xy") %>% llply(function(v) v^2) %>%
    add %>% imsplit("c") %>% add 
}

# detect.edges(subim(load.image(filenames[75]), y > 16)) %>% sqrt %>% plot
# 
# plot(subim(load.image(filenames[41]), y > 16))
# hazeFactorExampleFog   <- GetHorizAvgTrans(subim(load.image(filenames[41]), y > 16)) %>%
#   plot(, main = "Horizontal averaged transmission - Fog condition")
# plot(subim(load.image(filenames[58]), y > 16))
# hazeFactorExampleNoFog <- GetHorizAvgTrans(subim(load.image(filenames[58]), y > 16)) %>%
#   plot(, main = "Horizontal averaged transmission - No fog condition")

configDF <- read.csv("properties.csv")

location <- 1

filenames <- FilterNightTime(0, configDF, locationID = location)

filenames<-unlist(filenames)


fileStringPattern <- configDF[configDF$locationID == location,]$filePattern

print(length(filenames))


imageSummary <- foreach(file = iter(filenames), .combine = rbind) %dopar% {
  fileInformation <- FileNameParser(file, fileStringPattern)
  im <- subim(load.image(file), y > 16) #[, -(1 :17), ,]
  #imGradient <- get_gradient(im, "xy", scheme = 2L)
  data.table(name = fileInformation$name,
             dateTime = fileInformation$dateTime,
             meanEdge = detect.edges(im, 3) %>% sqrt %>% mean,
             changePoint = cpts(cpt.mean(GetHorizAvgTrans(im), penalty = "None"))
  )
  #hazeFactor <- GetHorizAvgTrans(im)
  #print(hazeFactor)
}

Load sensor values

sensorFiles <- list.files("../visDec/inst/extdata/Sensor",
                          pattern=glob2rx("MOR_DeBilt*.txt"),
                          full.names=TRUE)
sensorData <- ReadMORSensorData(sensorFiles)
setkey(sensorData, dateTime)
setkey(imageSummary, dateTime)
imageSummary <- merge(imageSummary, sensorData)
imageSummary[, medianMOR := median(c(FS261, TMM261, FS260), na.rm=TRUE), by = dateTime]
# imageSummary %>% .[ ,medianMOR := median(c(FS261, TMM261, FS260), na.rm =TRUE), by = dateTime]

Scatter plots

imageSummary[, visibility := factor(2000)]
imageSummary[medianMOR < 1000, visibility := factor(1000)]
imageSummary[medianMOR < 500, visibility := factor(500)]
imageSummary[medianMOR < 200, visibility := factor(200)]
#imageSummary[, transmission := read.csv("~/Dropbox/SWI_FogDetection/TrainingSet/Values.csv", sep=";")[, 3]]
# ggplot(imageSummary, aes(x = meanEdge, y = medianMOR, col = visibility)) + geom_point()
# ggplot(imageSummary[hour %in% seq(7, 16, by = 1), ], aes(x = log(medianMOR), y = meanEdge, col = visibility)) + geom_point()
# ggplot(imageSummary[hour %in% seq(7, 16, by = 1), ], aes(x = log(FS261), y = meanEdge, col = visibility)) + geom_point() + geom_quantile(aes(col=NULL), quantiles=0.5)
# ggplot(imageSummary[hour %in% seq(7, 16, by = 1), ], aes(x = log(medianMOR), y = meanEdge, col = visibility)) + geom_point() + geom_quantile(aes(col=NULL), quantiles=0.5)
ggplot(imageSummary[hour %in% seq(7, 16, by = 1), ], aes(x = log(medianMOR), y = changePoint, col = visibility)) + geom_point() + geom_quantile(aes(col=NULL), quantiles=0.5)