I’ve waxed lyrical about the Ricoh GR being a great UAV camera before - well the long awaited successor has been announced (but hasn’t yet landed) and summarised over at DPReview. The interesting aspects of the uprated specs are IIS (inbody image stabilisation), 24MP sensor and touchscreen. The resolution boost and IIS will be of significant interest to UAV users so it will be interesting to see how it performs out in the field.
I commented late last year about DJI becoming a camera manufacturer… this is an interesting and exciting move simply because the quality of imagery from drones has lagged significantly behind the rest of the camera industry. So whilst there are clearly regulatory challenges that need to be overcome through hardware and software engineering, the end-user is interested in the best visual imagery possible and drone manufacturers are starting to wake up to this fact.
So I was surprised to see that SeneseFly had announced back in 2016 its (then new) eBee Plus offered an hour of flight time within a ~1kg airframe that can incorporate RTK/PPK positioning (forget those ground control points!) and multi-spectral/thermal. But… they have also introduced their own homegrown air camera dubbed SODA (Sensor Optimised for Drone Applications). It is extremely light on specification however their website notes its a 20MP 1” fixed lens camera. This that likely utilises Sony’s standard 1” sensor, although it would be useful to clarify the exact size. Is it genuinely 13.2x8.8mm? Taking their GSD example figures, that breaks down to a 10mm focal length lens, equivalent 27mm on a full-frame camera. Interesting there is a single global electronic shutter which is a good thing for a fixed-wing aircraft: it should stop the problem of a rolling-shutter.
Fascinating to see the drone-camera market develop and it’d be great to see some results from this baby.
Earth-i has just launched a new SSTL built satellite which is claimed to be the first to provide full colour UHD video - UHD is 3840x2160 pixels (8MP), shooting at 50 fps (compared to Landsat 9’s ~12,000x12,000). That’s a lot of downstream data, although it would appear it’s not the video they’re interested in, but the multi-temporal data. Think super resolution to give them an effective ~70cm pixel size, but also stereo (and so 3D). This is the first of a planned 15 satellite constellation which could provide global coverage and much more agile mapping capabilities. Video is clearly the new high resolution!
DPReview report report on DxOMark’s tests of the Zenmuse X7 and it makes for some impressive reading… it’s a quality sensor that has nearly 14 stops of dynamic range with good low-light performance. This line from the review pretty much sums things up:
it delivers results that compete closely with those from a high-scoring APS-C format DSLR, despite being housed in a camera that’s mounted in a stabilized gimbal and specifically designed for aerial photography.
A couple of nice space links…. first a celebration of the Russian Soyuz system. An engineering marvel that is reliable and low cost - a feat for any product but the fact that this is over 50 years old and just keeps working is remarable.
One of the big drives for DJI has been the film industry and so there has been some buzz on the wires with the announcement of the Zenmuse X7, a camera built by DJI rather than using a partner’s system that has some interesting specs (see DP Review). This is primarily aimed at cinematographers, but as DP Review note this is actually a highly disruptive move by the company. It marks their entry into camera manufacture and introduces an APS-C sensor (24 MP), with a new lens mount and suite of lenses. So a larger sensor size but at a dramatically lower weight and smaller dimensions - the flange distance is a tiny 16.84mm with a minimum weight (including lens) of 630g.
So, let’s say it, this is a camera (and integrated into a system) that is destined for photogrammetry. Watch this space, disruption is coming!
So it’s no surprise to see Canon looking to move into this space and the (just rolling off the tongue) PD6E2000-AW-CJ1 is just that, as reported by DPReview. The drone itself is produced by Prodrone, a company Canon has invested in, and is not too dissimilar to the DJI Matrice 600, but kitted out with Canon’s ME20F-SH. This is firmly targeted at the disaster relief sector - the camera has an ISO of 4 million (yes, you read that right!) which will allow it to capture video in near-darkness.
The drone space continues to accelerate in terms of innovation, so expect to see an interesting and exciting roadmap of products appearing at breakneck speed!
Is this the next Uber Drone, coming to some skies near you? As the article says, would like to see 1000 hours of safe flight time first and the ability to fly with only two motors by feathering then should there be a failure. Exciting times though - I get a sense of scenes from The Fifth Element coming true (that would be Ruby Rhod first maybe)!!
And the cost of hiring a UAV is being driven down…. Dronebase is your AirBnB of UAV operators. Don’t pay over the top and go to one site to find them. The growth has been extremely rapid - see what their investors think. Its a great idea and, well, a very useful resource.
Way back in 2009 I published a paper on the Cookie Cutter which outlined a method (and accompanying script) for calculating the volume of drumlins. This worked in ArcGIS 9.2 using the Python interface to a number of ArcGIS toolbox tools. Fast forward 9 years since I first wrote the script and, not too surprisingly, it doesn’t work (thanks for telling me Arturs!).
I finally sat down a few weeks ago to bug fix the script which was actually easier than I thought it would. It’s actually comprised of two scripts - the first sets up some working directories and takes an input shapefile, splitting into a number of new shapelines (one per drumlin). The second script then performs the volume calculation on each drumlins. It turns out (given Im pretty much only calling Toolbox tools) that there wasn’t much to fix… a third party script splitting the initial shapefile had to be removed, a bug in the command adding a new field and then reference to ArcGIS 10.4 paths. For those wanting to use it, please download the attached files and follow the notes below.
I use WinPython 2.7 and then the excellent Spyder IDE to run the scripts from
in Spyder you need to change the Python console to the path of the one that ArcGIS has installed. Goto Tools -> Preferences -> Console -> Advanced Settings then change “Use the following Python Interpreter”. It should be something like: C:Python27ArcGIS10.4python.exe
at the top of the cookie_setup script set the project_directory to the location of the main input shapefile. For outlines, set the name of the input shapefile
Press F5 to run the setup script, creating the working directories and adding a new field to the shapefile
the next part needs to be performed manually (I haven’t had time to add in and test the Toolbox call)… add a new text field to the outlines attribute table called “split” and consecutively number each row from A1 to An (ie your last row). In QGIS the expression in the field calculator is concat(‘A’, @row_number )
save the file then use the ArcGIS Toolbox tool Analysis Tools->Extract->Split
use this to split the outlines shapefile based upon the “split” field you just created. Specify the “Target Workspace” as the “input” directory that has been created in your project directory
now load Cookie_Cutter into Spyder and again specify the following 5 inputs:
project_directory : the project directory
nextmap_dsm_img : the input DEM
gp.cellSize : the DEM cellsize (in metres)
tension_parameter : leave this as it is
buffer_parameter : the distance to buffer your drumlins (in metres). The example shows 20m, for a 5m DEM
on line 66-68 you might need to change the path to the listed toolboxes. This is specified for 10.4 at the moment
RUN IT! The console pane in Spyder should show you a whole load of information as it processes each drumlin. There will be a counter showing you which drumlin you are on
The key output is the Volume_Master.dbf table. You can open this in excel. It is zonal stats from ArcGIS for each individual drumlin (subtracted from the cookie cut DEM). The critical value is the SUM column that shows the total height for all pixels within the drumlin. Multiply this by 25 (for a 5x5 pixel) to give you drumlin volume.
UPDATE: If you can’t (or don’t want to) use Spyder you can just run the py script directly from the command line using the interpreter that ships with ArcGIS.
We’ve long been able to view remote video footage from a remote camera - when I was developing the kite aerial photography workflow I currently use, I experimented with a small spy camera that transmitted video footage back to an analogue LCD TV. It worked quite well and the intention was to use it to determine camera attitude and so allow use to remotely rotate the camera. Except for one BIG problem - when you are imaging the natural environment from relatively close range (60m), one piece of grass looks EXACTLY like another!!! In the end it was a pointless exercise.
Anyway, things move on and the next obvious way to integrate video was via AR (augmented reality) through a video overly in glasses. And true to form, DJI have announced another tie up, this time with Epson and their Moverio BT-300. See the announcement and some commentary on it over at DPReview. For consumer drones these make more sense as you tend to be viewing obliquely from (very) close range so positioning is important. Of course the glasses cost as much as the drone (!) but expect these products to become more prevalent.
Neither DEM is actually completely global missing, to greater or lesser extents, the polar regions. GDEM has greater coverage (owing to the overpass of the satellite) but is generally considered to be less accurate than SRTM. So its inevitable that a merged product should arrive which covers ~91% of the globe and mitigates much of the noise and void problems of the individual products. Full methodology can be read here.
1. Landsat 8: the base imagery they Google fall back to if no high res aerial imagery is available is now Landsat 8. Previously it was Landsat 7. Google need moderately up-to-date imagery but the Landsat 7 partial failure meant a slower process for acquiring global imagery. With enough global data in the back pocket, they can now switch to the newer sensor.
2. Google Earth Engine: Google’s cloud based image processing service is used to process the imagery. This leverages huge online image archives and processing power to allow very fast MASSIVE processing. So, in this case, for every pixel across the whole planet, using every available Landsat 8 image, select the one with the lowest cloud cover. Very neat way to create a global cloud free image. Or to put it in their marketing speak:
Like our previous mosaic, we mined data from nearly a petabyte of Landsat imagery-that’s more than 700 trillion individual pixels-to choose the best cloud-free pixels. To put that in perspective, 700 trillion pixels is 7,000 times more pixels than the estimated number of stars in the Milky Way Galaxy, or 70 times more pixels than the estimated number of galaxies in the Universe.
3. The Verge: just to show how mainstream this is, even The Verge picked up on it. Remote sensing is joe bloggs interesting now!
DPReview have a review of Bentley’s rather impressive 53 billion pixel of one of their cars on the Golden Gate Bridge. It might not be obvious at first glance but it is there!! Made up of over 700 photos taken at varying focal lengths (between 300 and 1500mm… bet that puppy cost a few pennies!) using a motorised panoramic tripod head it is one monster of an image that require a fare bit of TLC is post production - most notably the bridge can move ~8m in either direction in high winds! Head on over to Bentley to take a peek.
DPReview report on a new DJI/Hasselblad tie up that sees Hasselblad bundle their A5D (see my earlier post) medium format camera with DJI’s industrial Matrice 600 drone. The drone has a flight time of ~20mins with the 6kg payload, whilst the A5D ranges up to 60 MP and weighs 1.3kg (body only). This is clearly targeted at the professional aerial imaging sector so expect to see more tie ups on this front.
I’ve not blogged on it yet, but Hasselblad have also recently announced their new X1D, the first mirrorless medium format camera which has a 50MP (43.8x32.9mm) sensor but only weighs 725g (body). Expect to these this making its way into drones pretty shortly! Well… if you have the £7,188 asking price to hand.
An interesting article over at Amateur Photographer which picks up an Olympus press release about the development of an RGB/NIR sensor for use in consumer grade cameras. The use of digital cameras for NIR imaging (e.g. my dead leaf photo) has been common for many years and is achieved by having a longer exposure (as the sensor is less sensitive to NIR) and placing a NIR cut filter in front of the lens (e.g. Hoya 720). Specialists such as Advanced Camera Services will even convert your camera to IR by removing the internal IR filter. Sensefly use a modified Canon S110 for the eBee UAV which can image in RGB, NIR or red edge. Which is why the Olympus announcement is interesting (for the light weight/low cost UAV sector) as I’m not aware of a major manufacturer developing a single sensor for imaging 4 bands. A traditional approach is to use a bayer array over a sensor sensitive to RGB and then interpolate (demosaic) the image to three RGB layers. Olympus appear to have extended this to 4 bands by developing realtime demosaicing to support it. The sensor is probably a standard one, albeit perhaps more sensitive to NIR. Lead time could be awhile as this is in development but it clearly shows the direction of travel.