My project is still in the nebulous planning (can I afford it, can I build it) phase and I've noodled out alot of info on air frames and autonomous flight but I'm missing an important part of the puzzle. My goal is to fly lawnmower patterns over ~90 acres for mosaic mapping and possible wildlife spotting. With this as the goal, I'm wondering about flight altitude vs camera resolution. I know that if I fly higher (300m vs 100m), I'll need a higher resolution to get the same quality image. I don't think I'm saying that right but I bet you get my point.
So, are there any calculators or trial and error experience that says for example: 7 Mega Pixels is good for 100m altitude stills if I want 4cm/per pixel resolution (pulled this number out of thin air as I don't know what is good enough yet)? How does that scale if I want to fly at 300m? 9MP camera? I guess I'm asking, for a given altitude, and given pixel=X centimeters, what resolution camera do I need? Is it more complex than this? Lens diameter?
On the fantasy end of this project, the future may bring low light or infrared cameras into the mix. Has anyone flown them? Good results? Thoughts?
Replies
Joe, I used to be S Africa, and I still have many connections there, but I now live in Kansas. The tortoises are in scrubland in western Kansas that looks a lot like Karoo scrub. As Peter pointed out, the cold-blooded tortoises would not give a thermal signature, so we will have to use visible imaging. It will be VERY challenging, but we won't know for sure until we try. I'm thinking of flying low, slow transects with a hexcopter. Bison should be relatively easy - like counting cattle!
Do you know Cornelis van der Waal from Agri-Ecological Services in Omaruru? He is likely to have an interest in what you do.
My primary targets, at least initially, will be bison on open grassland - probably one of the easiest targets for counting compared to what you are doing in the Namibian bushveld. I'm also thinking about surveying tortoises in dry, Karoo-like scrubland. That may turn out to be a major challenge!
Another question: have you tried to count animals using HD video? My thoughts are that you will have low resolution, but you may gain information from the changing perspective as the airplane moves over the target animals.
Thank you for sharing your experience in counting wildilfe from the air. It is very useful information to many of us. I have a question about the effects of UAVs flying at low altitudes (such as 100 to 200m) on animal behavior. I assume that each species will be different, but is there an optimal altitude where most animals will not take evasive action, or be distressed?
Also, what is the optimal time of day for this work? I assume that the middle of the day could be a problem due to animals seeking shade, while light quality could be a problem at other times of day.
Thanks much!
"So, are there any calculators"
Hi Kevin, you can download the mission planner GCS, and use the Camera tool, to make those calculations. its easy and free.
try to keep, your setup simple for starters. Maybe a thermal veiw of the game area using tau with an FPV setup. then you can procced to mapping. I suggest that, because you will be able to idendify the populations and regions and you want have to stich-combine-georefernce and analyse a map, rapidly. the 90% of mapping is on the ground side and its very very tricky, hour consuming and till now needs a lot of knowledge and software. Ofcourse that 10% (aircraft & cam) are the base.
@Joe,
I admire your work and the consistency with you persue profesionalism. your suggetions and arguments are always based on facts and field work. But this time, i found two key factors that i believe they made you to suggest these cams.
1st is the extraordinary ground speed for most of available airframes(20m/s), which results in a high interval speed (note. not shutter speed). So you limit down your options at superzooms or mirrorless cams.
2nd using an external shutter, you extend the chain sequence from klik -zoom focus-acquisition and add failure points. Chkd can extend any Canons possibilites so you can shut-focus-aquire and already preconfigure the shutter speed, iso, and other variables at a Known interval! also in our job, a very significant factor besides the spatial resolution is the bit depth, which raw (12bit vs jpeg 8bit) can offer, with CHKD even at low end P&S cams.
I have used many kinds of combos like yourself mainly for cadastral maps, forest cartography and agriculture. Now, in most of my works i use a light p&s canon (ixus) at 113m flying alt (legitimate ceiling) which gives me about 3cm/p! and 12bit raw! with 3s interval .
You can see samples of my work here
Hi again Kevin,
Just need to add and clear up a few points -
For recognition of animal species ( Buck versus big cat for example, NOT type within species, ie, a Kudu or a Gemsbok..), you need the item of interest to cover at least a pixel area of around 15 x 5pixels. This is the bare minimum and is strongly influenced by bush density, and contrast , ie, colour of animal versus background..To recognize the type within species you may need to go up to double the pixel area. Obviously knowing what animal types are in the area will narrow the choice and allow lower resolution recognition.
This also demands the best performance from the system - a reasonable stable flight platform ( the lighter the weight, the poorer the stability in mild wind, gusts and thermals, typical in any bush areas), LOW vibration - gas engine planes are a nightmare - you have to use a much bigger plane since you need to fit the camera in a heavy frame to give it inertial mass, and then shock-mount that frame to the A/C payload bay - I did that with GUPPY - check my blog- and the 300gram camera was fitted to a 500gram frame and then to very good shock mounts - only then were the pics acceptable - admittedly, it was a 100cc engine, BUT it was a twin boxer engine, and the airframe was around 20KG, which is a big inertial mass to absorb the vibration. Scale that down to a 10cc single cyl engine ( a jackhammer..) and a 3kg airframe and you are worse off. Electric is the way for smaller craft.
Also, i see lots of folk advocating the use of CHKDK or whatever its called for the cannon cams - I think its a waste of time.
My 'favourite' camera at the moment is the Panasonic Lumix GX1 - it is 16MP, which is low - I would very much like a 20Mpixel, but it is FAST, with a very good, low distortion 14mm lens ( 28mm equiv) and can do auto-focus-lock for pre-focus at infinity. Our autopilot controls the camera snap and focus and we normally use it as follows - at the start of a photo leg the focus is press and the camera takes a second or so to focus, then at the first photo point we activate snap and release ONLY snap, keeping focus active. On this camera that means the focus remains locked, and we just snap thereafter, at down to 0.3sec per snap. There are few cameras that will do this - The Samsung NX200 ( 22Mpixel) which we were using does not- the only reliable way with that camera was to fit a fully manual lens and pre-focus. The GX1 is very good. The loss due to 16Mpix versus 22Mpix is not huge,
for example, at 2cm GSD, the flight height is 74m for panasonic and 150m for Samsung, and for a 100hectare area flown at 4GSD, you fly around 15% more legs for the Panasonic versus Samsung. But the end result is the same - fly lower for better resolution. Also , the Panasonic has a smaller sensor size, and so the field of view is not greatly affected.
Here is a spreadsheet calculator for ground resolution versus height versus everything - you need ALL the camera specs - get them from DPREVIEW.com
Regards
Joe
CameraCalcsJoe.xls
Slightly off topic, here is my rather lame attempt at a calculating optimal camera specs, altitude and and airspeed for aerial photography. Inputs are in yellow: 35MM Equivalent Focal Length (in mm), Camera Resolution (in megapixels, assumes photos are taken in 4:3 format), Height above ground (in meters) and airspeed (in meters per second). Outputs are Picture Resolution (in centimeters per pixel or CMPP - not quite right at the moment I don't think but a good estimate), Required shutter frequency for optimal overlap (as per dronemapper.com guidelines http://dronemapper.com/guidelines) and Cross Track Gap (in meters - effectively the gap between the tracks of the lawnmower pattern).
Comment welcome.
Calculator Height.xlsx
Hi Kevin,
You are sort of asking the right questions, and I have been hanging in the background for a while to see what bright answer pop up...But no-one seems to be answering you!
I am doing exactly what you want to do so maybe I can help. We have a system called SurVoyeur ( see my blog on DIYDrones) and these are is use by land surveyors and as well by the Ministry of Conservation here. I can provide some info re both appilications.
Surveyors are really only interested in achieving a certain ground resolution - this is not the same as being able to recognize the observed feature.
To determine what type of camera ( how many Mpixels) and what type of lens is needed is simple math, but is also influenced by the speed at which you can fly, and the rate at which the camera can take and store photos. But for survey applications, the greatest influence is the software used to process the images. These software all require substantial lateral and forward overlap to be able to suitable stitch and process. In addition, MANY accurate ground placed GPS targets are needed ( tie- points) to 'tie' the ground features down to mother earth, else they rise and fall and the lie of the land is no longer accurate...
Typically the side to side overlap required is at least 50% with 60% preferred, and forward around 70 to 75%.
To know how often a photo has to be taken in the forward direction you have to know the camera field of view on the ground, from the height AGL that will give you the GSD required.
The angular field of view is determined from the lens focal length and the CCD Sensor size.
Lets take 3 cameras - A Samsung NX200 (22Mpixel) with a 21mm lens , a Panasonic GX1 (16Mpixel)with 14mm lens
and a RICOH D4 (10Mpixel) with a 6mm lens
The bold text gives the height you have to fly at to get the required GSD.
The max time between pics show how fast you need to be able to take and store each photo, ASSUMING(!) that the ground speed = airspeed. If you have a tail wind your ground speed is faster, so the rate of photos per time increases.
There are very few cameras that can take photos in full autofocus mode at a rate of 1 every 0.5 sec - that is approx the rate at a ground speed of 28m/s. We fly at 20m/s, and have flown in wind of 15m/s...
SO we operate the camera in either of two ways, depending on what modes the camera provides: either manual focus, ie, a true manual lens is fitted ( as is on the Samsung below - a Pentax 21mm manual lens), which we pre-focus at infinity, and operate the camera in aperture priority mode ( not the greatest, since the aperture is fixed then), or on the Panasonic GX1 we operate in Shutter Priority, set to a high shutter speed ( around 1/1600 second), but pre-focus the lens and press autofocus lock - that focuses at infinity, and set aperture automatically for light variation. The fast shutter ensures no blurring, and the fixed focus means very rapid photos - we get one every 0.3secs..
Then for 2cm GSD:
Samsung Pentax
Panasonic GX1
Ricoh D4
CCD Width
23.5
17.3
7.2
mm
CCD Height
15.7
13
5.3
mm
CCD X Resolution
5472.0
4592.0
3648.0
pixels
CCD Y Resolution
3648.0
3448.0
2736.0
pixels
Focal Length
21.0
14.0
6.0
mm
Required GSD
2.0
2.0
2.0
cm
Req Overlap X
50.0
50.0
50.0
%
Req Overlap Y
70.0
70.0
70.0
%
Angle of View - X
58.5
63.4
61.8
deg
Angle of View - Y
41.0
49.8
47.8
deg
Image Ground Area X
109.4
91.8
73.0
m
Image Ground Area y
73.0
69.0
54.7
m
Req Hagl X
97.8
74.3
61.0
m
Req Hagl Y
97.6
74.3
61.7
m
Distance between pics X
54.7
41.3
36.5
m
Distance between pics Y
21.9
20.7
16.4
m
Flying Speed
20.0
20.0
20.0
m/s
Max time between pics
1.1
1.0
0.8
s
and for 4cm GSD
SamsungPentax
Panasonic GX1
Ricoh D4
CCD Width
23.5
17.3
7.2
mm
CCD Height
15.7
13
5.3
mm
CCD X Resolution
5472.0
4592.0
3648.0
pixels
CCD Y Resolution
3648.0
3448.0
2736.0
pixels
Focal Length
21.0
14.0
6.0
mm
Required GSD
4.0
4.0
4.0
cm
Req Overlap X
50.0
50
50.0
%
Req Overlap Y
70.0
70.0
70.0
%
Angle of View - X
58.5
63.4
61.8
deg
Angle of View - Y
41.0
49.8
47.8
deg
Image Ground Area X
218.9
183.7
145.9
m
Image Ground Area y
145.9
137.9
109.4
m
Req Hagl X
195.6
148.6
121.9
m
Req Hagl Y
195.2
148.5
123.4
m
Distance between pics X
109.4
82.7
73.0
m
Distance between pics Y
43.8
41.4
32.8
m
Flying Speed
20.0
20.0
20.0
m/s
Max time between pics
2.2
2.1
1.6
s
All good and well for Survey use..
For animal spotting and tracking its a different story.
Animals are masters of disguise and MUCH better resolution is required to discern them from there natural backgrounds.
Elephant are easily seen, and there young easily counted from 150meters with a 10Mpixel camera, on open plains and savanna grasslands. At 300meters with a 10Mpixel camera, the young are detected with difficulty, if at all. If the contrast is low, the adults may be seen, normally only if they are in a group, as the pattern strikes the eye. A single elephant will not be seen in such a photo. In bushier terrain, Elephant are really only detectable from a max height of 150 meters with a 10Mpixel camera.
As a guide -
at 100m height and 10Mpixel you will detect and recognize a small deer size buck
at 200m you may detect but will not recognize the same buck, but will with a 20Mpixel camera.
At 100m with 10Mpixel height you may detect a jackal, but will not be able to see the type.
with 20Mpixel you will be able to discern the type if the contrast is good, but not at 150meters...
The biggest problem with game detection is that they are well camouflaged and stand still on a photo - there is no movement, which is what the eye is most sensitive to - that is why you will spot more animals from a chopper than from a photo - the movement makes the animal stand out from its habitat.
Game counting by this means is a waste of time - unless it is in the wide open plains with nowhere to hide, and where the animals have contrast to the background. In addition, in bushy terrain, the animals forage under bushes and trees, and a UAV does not scare them out - a chopper sends them running.
You choice of camera is important - it must be fast, it should have as wide angle lens as possible, with low distortion, it must be able to be controlled externally ( Focus and snap), it must preferably have an auto-focus-lock ( AFL) capability, etc. The larger (Physical) sized sensor cameras are not necessarily the better choice - the lenses end up MUCH larger and heavier.
We also fly one of the FLIR IR cameras ( video) - the TAU-2. That is another story - very good camera for hot body detection - don't let anyone fool you - using a clickety-click camera with the 'IR-filter' removed does not do the job - there is ABSOLUTELY no comparison.
Hope this helps - if you need more info please ask - we do a lot of this type of work here in Namibia.
Joe
The Nampilot
This camera Sony RX100 is all the rage lately for high MP (20.2MP) and HD recording. Plus a 3.6x Zoom compliments it nicely. It also is well below your 900.00 range and currently there are a few ppl working on control hacks and lens hacks.
Remember, MP is not the central point of good images, the cameras ability to collect light is equally important.
It didn't seem that this Moose had any problem with this UAV at about 5 meters!
Bullwinkle Gets spotted
Good luck!
I had a similar challenge: I wanted to use a forward looking infra-red (FLIR) camera,
which has notoriously big pixels and low resolution, to get a count of the white-tailed
deer population, which is increasing exponentially. We figured we would have to fly
QUIETLY, below 100 feet, in the fall, winter and spring when there are no leaves on
the trees, if we wanted to be able to acquire images that would allow us to tell the
difference between a buck, a doe and their offspring. Nothing else in the Hamilton,
Ontario area has a heat signature that matches a white-tailed deer.
-peter
What kind of animals are you looking for? That will help determine how good of a camera/lens you need. If you're looking for elephants that doesn't require as much resolution as if you're looking for squirrels.
-
1
-
2
of 2 Next