benchmarking tour parameters

[willmorrison1]

What tour parameters are best? How many images and across what interval? What resolution? row # is the region ID, rows are permuted parameters

intervalSize downwardImage nSamplesAroundOrigin imageRes

• [x] 1 100 FALSE 4 2000

• [x] 2 150 FALSE 4 2000

• [x] 3 100 TRUE 5 2000

• [x] 4 150 TRUE 5 2000

• [x] 5 100 FALSE 5 2000

• [x] 6 150 FALSE 5 2000

• [x] 7 100 FALSE 4 3800

• [x] 8 150 FALSE 4 3800

• [x] 9 100 TRUE 5 3800

• [x] 10 150 TRUE 5 3800

• [x] 11 100 FALSE 5 3800

• [x] 12 150 FALSE 5 3800

[dpoursanidis]

On the imageRes, the higher the better.

The rest, depending on the quality in the Google Earth and the application the results will be used for.

[willmorrison1]

Thanks Dimitris. Agreed. The above is a brain dump and work in progress evaluation!

I am hoping to get away with lower res, else I anticipate requiring a lot of storage space. Given the limited detail of the GE model, I will try out all of these permutations across a complex high rise area and evaluate using the highest resolution result.

[dpoursanidis]

Shall we compare GE results againts an rgb drone-based result, if that is possible?

[willmorrison1]

Sounds like a very good idea. I was just looking to see if Heraklion has GE 3D data and it does, so this comparison is possible. Can you send a shape file or some coordinates of the drone data spatial extent?

[dpoursanidis]

I will prepare and send soon. Can you send a reminder over mail on Thursday?

[willmorrison1]

updated advised configuration for tough areas e.g. dense high rise https://github.com/kitbenjamin/googleEarthImageCollection/commit/88ce8c2533ce92cf5562eba9e58ac66fd18adeb6. benchmarking will be quite qualitative (no time for in-depth analysis at the moment)

[willmorrison1]

qualitative results: with

9 100 TRUE 5 3800

the benchmark (smallest interval length, highest img res, most images per interval)

 10 150 TRUE 5 3800

is a reasonable image sample density (interval length of 150 m vs 100 m, 5 images per interval)

[willmorrison1]

image collection time/storage space requirement back of envelope calculations turned into a quick script:

for reference/update later:

library(dplyr)
library(cowplot)
library(QOLfunctions)
dfOut <- expand.grid(regionXY_m = seq(10000, 15000, by = 5), 
                     interval_m = seq(150, 250, by = 1), 
                     imagePerinterval_m = 5)
maxImg <- 100000
minImg <- 1000
dfOut$totalImg <- with(dfOut, (ceiling(regionXY_m / interval_m)^2) * imagePerinterval_m)
dfOut$totalImg <- mround(dfOut$totalImg, 5000)
dfOut <- dfOut %>%
  dplyr::filter(between(totalImg, minImg, maxImg))

#approx 1000 images per hour
timePerImg_s <- (60*60) / 1000
#approx 11GB per 1000 images
spacePerImg <- 12

basePlot <- ggplot(dfOut, aes(regionXY_m, interval_m)) + 
  # scale_fill_distiller(palette = "Spectral") + 
  xlab("Square region length (m)") + 
  ylab("Image interval (m)") +
  theme_bw() +
  scale_x_continuous(guide = guide_axis(check.overlap = TRUE))

imageNo <- basePlot +
  geom_raster(aes(fill = as.factor(totalImg))) +
  labs(fill = "n#") +
  ggtitle("Number of images")

imageTime <- basePlot +
  geom_raster(aes(fill = as.factor((totalImg * timePerImg_s) / 60 / 60))) +
  labs(fill = "days") +
  ggtitle("Colletion time") 

imageSpace <-  basePlot +
  geom_raster(aes(fill = as.factor((totalImg * spacePerImg) / 1000 / 1000))) +
  labs(fill = "TB") +
  ggtitle("Storage space required") 

pltOut <- cowplot::plot_grid(imageNo, imageTime, imageSpace, ncol = 1, align = "hv")

ggsave(filename = "C:/Users/micromet/Desktop/regionStats.png", 
       plot = pltOut, height = 7, width = 4, dpi = 600)

[willmorrison1]

tried 200 m image interval and is too coarse. 150 m is recommended image interval so long as it remains a constant

[willmorrison1]

large area image collection (9 x 9 km) stats

Summary:

• 2 days to get 9 x 9 km area of images
• still some bugs to crush

dir <- "D:/GE/ImageCollection/regions/Central_London/googleEarthOut"

allFiles <- list.files(dir, recursive = TRUE, pattern = "movie", full.names = TRUE)
allFiles_info <- file.info(allFiles)
totalTime_s <- as.numeric(max(allFiles_info$ctime) - min(allFiles_info$ctime)) * 24 * 60 * 60
print(totalTime_s)
#> [1] 126902.5
plot(diff(sort(allFiles_info$atime)), ylim = c(0, 3), pch = 20, cex = 0.1)
abline(h = median(diff(sort(allFiles_info$atime))), col = "red")

^{Created on 2020-06-08 by the reprex package (v0.3.0)}