Resolution Tests for MU UAV Imagery

In preparation for Mansfield’s new imaging UAV, I performed some tests with our primary imaging camera today to determine the potential resolution of the sensor, spatial coverage, and its viability for various imaging applications. In estimating the potential power of our sensors, it appears we might be capable of mapping on a scale that is largely unseen (“micro-mapping”), with the drone potentially capable of providing very high resolution remotely sensed imagery. However, testing was needed to confirm such possibility before application of the technology in the field.

Equipment

Image from DPReview.comThe camera we will be using with the drone is a Panasonic DMC-GH2, a mid-level “prosumer” digital SLR recommended by Quadrocopter for this application.  The camera is equipped with a 16.05 megapixel sensor, and a 14-42mm (f:3.6-5.4) kit lens.  Because the lens focal length can only be changed manually– meaning once the drone is in the air, our focal length is chosen—both finite settings (14mm and 42mm) were tested.  The pictures were taken in RAW mode, the highest quality digital file that the camera is capable of producing, and processed using Adobe Photoshop CS6. Here, to ensure compatibility with the web, they have been converted to JPG format.

 

Test 1: Field of View and Coverage

The drone will allow for flights at a controlled altitude above the surface.  To determine potential coverage of the camera from various altitudes, I took some simple test shots and used basic trigonometric functions to determine the camera’s field of view (see Figures 1-4, below).

Field of View at 14mm

From a distance of six feet (72”, 182.88cm), the camera covered an area that was precisely six feet (72”, 182.88cm) wide, and four feet (48”, 121.92cm) long, for a total coverage of 24ft2 (3456 square inches, 2.2297m2).

 

 

Figure 1 (left) and 2 (right): Calculating field of view with 14mm lens. The field of view for this camera/lens combination is 53.13° x 36.87° at 14mm focal length.

 

Field of View at 42mm

From a distance of six feet (72”, 182.88cm), the camera covered an area that was precisely 2.5 feet (30”, 76.2cm) wide, and 1.875 feet (22.5”, 57.15cm) wide, for a total coverage of 4.6875ft2 (675 square inches, 0.4355m2).  At 42mm, the lens shows 19.53% of the coverage area of the same lens at 14mm.

 

 

Figure 3 (left) and 4 (right): Calculating field of view with 14mm lens. The field of view for this camera/lens combination is 23.526° x 17.762° at 42 mm focal length.
With these measurements for field of view, it’s pretty easy to calculate coverage from various distances and hence, various altitudes of the drone in flight, using the same trigonometric equation, shown in general terms in Figure 5.

Figure 5: General Calculation for Coverage, Where D is Distance, 2X is Field of View (FOV), and C is Spatial Coverage.

For the purposes of this test, I used five foot intervals starting with five feet, up to 25 feet, just to give an idea of our capabilities for high-resolution imagery. I simply divided the number of pixels for each dimension by the coverage of that dimension to determine this. Table 1 below shows the results of these calculations.

Table 1: Potential Spatial Coverage at 14mm and 42mm Focal Lengths, Five Foot Distance Intervals

 

Distance 14mm Focal Length 42mm Focal Length
5 feet (1.524m) 5′ x 3.33′ = 16.67ft2
(1.524m x 1.016m = 1.548m2)
2.083′ x 1.563′ = 3.26ft2
(0.635m x 0.476m = 0.302m2)
10 feet (3.048m) 10′ x 6.67′ = 66.67ft2
(3.048m x 2.034m = 6.12m2)
4.167′ x 3.125′ = 13.02ft2
(1.27m x 0.953m = 1.21m2)
15 feet (4.572m) 15′ x 10′ = 150ft2
(4.572m x 3.048m = 13.935m2)
6.25′ x 4.688′ = 29.3ft2
(1.905m x 1.429m = 2.72m2)
20 feet (6.096m) 20′ x 13.33′ = 266.67ft2
(6.096m x 4.064m = 24.774m2)
8.33′ x 6.25′ = 52.06ft2
(2.539m x 1.905m = 4.838m2)
25 feet (7.62m) 25′ x 16.67′ = 416.67ft2
(7.62m x 5.08m = 38.71m2)
10.42′ x 7.813′ = 81.41ft2
(3.176m x 2.381m = 7.562m2)

Determining the coverage at each altitude is also important for understanding the feasibility of the drone for various mapping applications. Because the drone’s power is limited to approximately 75 minutes of daily flight between charges, and because limited coverage for each image means more work hours to stitch imagery, these numbers will help us determine exactly what kind of detail in larger-scale projects is possible.

 

Test 2: Resolution

While the camera is rated as having a 16.05 megapixel sensor, generally the output resolution of digital files from any digital camera are slightly different than the sensor size.  For this camera, the actual resolution of the output imagery in RAW format is 4608 x 3456 pixels, for an actual resolution of 15,925,248 pixels per image (15.925 megapixels).

With this information, a theoretical spatial resolution for this sensor can be determined, dependent upon distance from the subject and lens focal length, as shown in Table 2:
Table 2: Theoretical Spatial Resolution, Based on Image Size and Distance/Coverage Calculations for 14mm and 42mm Focal Length

 

Distance 14mm Focal Length 42mm Focal Length
5 feet (1.524m) 0.013″ per pixel
(0.33mm per pixel)
0.0054″ per pixel
(0.138mm per pixel)
10 feet (3.048m) 0.025″ per pixel
(0.66mm per pixel)
0.0109″ per pixel
(0.275mm per pixel)
15 feet (4.572m) 0.039″ per pixel
(0.99mm per pixel)
0.0163″ per pixel
(0.413mm per pixel)
20 feet (6.096m) 0.052″ per pixel
(1.32mm per pixel)
0.0217″ per pixel
(0.551mm per pixel)
25 feet (7.62m) 0.065″ per pixel
(1.65mm per pixel)
0.0271″ per pixel
(0.689mm per pixel)

Imaging at this resolution could contribute to micro-mapping, providing data that of far greater detail than what most cartographers can access. However, because sensor and optical limitations are evident beyond theoretically calculable pixel densities in any digital photography, these resolutions are strictly theoretical and needed to be tested with a more practical, real-world application.

To do this, I took the camera outside with two charts for determining grain size of sedimentary particles from Walker & Cohen’s (2009) collection of field resources, The Geoscience Handbook. The two charts were identical, except that one identified light-colored particles on a black field, while the other identified dark-colored particles on a white field. Also important were the 10mm grids below the charts, which provided the basis for calculating the resolution of each image:

Figure 6: Walker & Cohen’s Grain Size Charts (2009: 159-160)

I shot photos of the chart from distances in five foot intervals, using both a 14mm and 42mm focal length.  The shots were taken at approximately noon on an overcast, late autumn day and on a shaded northwestern wall of the building, so the light was quite soft.

The coverage provided at each distance and length was pretty much as expected, though the 42mm pictures below show up darker because the lens has a variable aperture that allows less light in at higher focal lengths. To ensure as few variables as possible were changed, I programmed the camera to maintain the same shutter speed (1/1000) and ISO (400) despite the difference in aperture.

Table 3: Full Imagery at 14mm and 42mm Focal Length, from Five Foot Intervals of Distance

 

Distance 14mm Focal Length 42mm Focal Length
5 feet (1.524m)
10 feet (3.048m)
15 feet (4.572m)
20 feet (6.096m)
25 feet (7.62m)

 

 

Resolution Test Results

I brought the imagery into Adobe Photoshop and cropped it to a few different widths, doing my best to maintain a standard width to the crop. The images below show the basic results of the test, but please note that these particular images are shown for illustrative, not analytic, purposes. [All analysis was completed with the original RAW imagery, which is not web-friendly]. With each pair of images, I’ve included the calculated spatial resolution, plus a brief discussion of practical application of this resolution in terms of which particle diameters were identifiable as individuals:

14mm Focal Length at Five Feet


Figure 6: Calculated Resolution of 0.0164″ per pixel (0.427mm per pixel) at 14mm Focal Length, 5ft Distance. Result is 0.097mm per pixel lower than calculated theoretical resolution (Table 2).

Figures 7 & 8: Grain Particle Sizes at 14mm Focal Length, 5ft Distance. At this distance and focal length, 100% of particles with a diameter of 1.0mm were identifiable on the light particle chart, while approximately 95% of the dark particles where identifiable individually. That percentage dropped off dramatically at a diameter of 0.5mm, with approximately 30% of particles of that size differentiated on both charts.

 

42mm Focal Length at Five Feet


Figure 9: Calculated Resolution of 0.006″ per Pixel (0.15mm per Pixel) at 42mm Focal Length, 5ft Distance. Result is 0.028mm per pixel lower than calculated theoretical resolution (Table 2).



Figures 10 & 11: Grain Particle Sizes at 42mm Focal Length, 5ft Distance. Unsurprisingly, this combination yielded the highest resolution of all test shots. At this resolution, 100% of the particles were identifiable on an individual basis at a 0.5mm diameter on both the dark and white particle chart. In addition, approximately 90% of the particles at d=0.25mm and an astonishing 30% of the particles with a diameter of 0.1mm were identifiable at this resolution.

 

14mm Focal Length at Ten Feet


Figure 12: Calculated Resolution of 0.03″ per Pixel (0.77mm per Pixel) at 14mm Focal Length, 10ft Distance. Result is 0.11mm per pixel lower than calculated theoretical resolution (Table 2).

Figures 13 & 14: Grain Particle Sizes at 14mm Focal Length, 10ft Distance. At this distance and focal length, all grain sizes of 1.0mm and below are indistinguishable, with variation of texture barely noticeable at 1.0mm. Some 90% of particles were individually identifiable at a diameter of 2.0mm, while even the larger grain sizes lacked 100% precision.

 

42mm Focal Length at Ten Feet


Figure 15: Calculated Resolution of 0.012″ per Pixel (0.294mm per Pixel) at 42mm Focal Length, 10ft Distance. Result is 0.019mm per pixel lower than calculated theoretical resolution (Table 2).


Figures 16 & 17: Grain Particle Sizes at 42mm Focal Length, 10ft Distance. At this distance and focal length, 75% of grains at 0.5mm diameter were individually identifiable, while 100% of grains 1.0mm and above were identifiable. Textures were noticeable at 0.25mm diameter, but individual grains were impossible to distinguish.

 

14mm Focal Length at 15 Feet


Figure 18: Calculated Resolution of 0.049″ per Pixel (1.25mm per Pixel) at 14mm Focal Length, 15ft Distance. Result is 0.26mm per pixel lower than calculated theoretical resolution (Table 2).


Figures 19 & 20: Grain Particle Sizes at 14mm Focal Length, 15ft Distance. At 15 feet, the resolution of the sensor using a 14mm focal length really begins to drop off. Approximately 30% of all particles with 2.0mm diameter or larger are individually identifiable, while 1.0mm diameter grains are only visible as a fuzzy texture. Not even the grains with a diameter of 7.0mm are clear at this distance.

 

42mm Focal Length at 15 Feet


Figure 21: Calculated Resolution of 0.017″ per Pixel (0.427mm per Pixel) at 42mm Focal Length, 15ft Distance. Result is 0.014mm per pixel lower than calculated theoretical resolution (Table 2).


Figures 22 & 23: Grain Particle Sizes at 42mm Focal Length, 15ft Distance. To me, this was the most impressive result. At a distance of 15 feet, with a 42mm focal length, fully 100% of the 1.0mm and larger diameter grains remained individually identifiable, while 0.5mm grains show as a textured surface.

 

14mm Focal Length at 20 Feet


Figure 24: Calculated Resolution of 0.066″ per Pixel (1.67mm per Pixel) at 14mm Focal Length, 20ft Distance. Result is 0.35mm per pixel lower than calculated theoretical resolution (Table 2).


Figures 25 & 26: Grain Particle Sizes at 14mm Focal Length, 20ft Distance. At 20 feet distance and 14mm focal length, most of the particles are indistinguishable. Grains with a diameter of 3.0mm or less only appear as textured surfaces, and only perhaps 70% of grains with a diameter of 5.0mm are identifiable. Even the grains with a diameter of 7.0mm are fuzzy and blurred together.

 

42mm Focal Length at 20 Feet


Figure 27: Calculated Resolution of 0.023″ per Pixel (0.588mm per Pixel) at 42mm Focal Length, 20ft Distance. Result is 0.037mm per pixel lower than calculated theoretical resolution (Table 2).


Figures 28 & 29: Grain Particle Sizes at 42mm Focal Length, 20ft Distance. Remarkably, at 20 feet, using a 42mm focal length allows the 1.0mm particles to still be largely visible, perhaps 40% of them individually identifiable. Smaller particles, such as 0.5mm, show up only as a textured surface, while all particles larger than 1.0mm show up relatively clearly.

 

14mm Focal Length at 25 Feet


Figure 30: Calculated Resolution of 0.081″ per Pixel (2.07mm per Pixel) at 14mm Focal Length, 25ft Distance. Result is 0.42mm per pixel lower than calculated theoretical resolution (Table 2).


Figures 31 & 32: Grain Particle Sizes at 14mm Focal Length, 25ft Distance. At 25 feet, the resolution of our camera at 14mm focal length is getting much lower. All grain sizes with a diameter of 3.0mm or less appear to be textured surfaces, while 5.0mm grains are largely indistinguishable. The 7.0mm diameter grains are somewhat identifiable, but difficult to separate.

 

42mm Focal Length at 25 Feet


Figure 33: Calculated Resolution of 0.029″ per Pixel (0.73mm per Pixel) at 42mm Focal Length, 25ft Distance. Result is 0.041mm per pixel lower than calculated theoretical resolution (Table 2).


Figures 34 & 35: Grain Particle Sizes at 42mm Focal Length, 25ft Distance. At 25 feet, using a 42mm focal length still produces impressive results. While the 1.0mm grains now largely look like a textured surface, the larger sizes generally exceed 90% of grains being individually distinguishable.

 

Summary and Discussion

Resolutions captured in the test imagery were largely inline with expectations, with most images showing resolutions quite close to the theoretical calculations. In most cases, the differences were small enough that they could have been the result of measuring error.

The difference between hypothetical resolution (from Table 2) and calculated resolution from test imagery is shown below in Table 4 (14mm focal length) and Table 5 (42mm focal length).

Table 4: Summary of Calculated Resolution from Actual Imagery Compared to Hypothetical Resolution, 14mm Focal Length

 

Distance 14mm Hypoth 14mm Actual Difference
5 feet (1.524m) 0.013″ per pixel
(0.33mm per pixel)
0.0164″ per pixel
(0.427mm per pixel)
-0.0034″ per pixel
(-0.097mm per pixel)
10 feet (3.048m) 0.025″ per pixel
(0.66mm per pixel)
0.03″ per pixel
(0.77mm per pixel)
-0.005″ per pixel
(-0.11mm per pixel)
15 feet (4.572m) 0.039″ per pixel
(0.99mm per pixel)
0.049″ per pixel
(1.25mm per pixel)
-0.01″ per pixel
(-0.26mm per pixel)
20 feet (6.096m) 0.052″ per pixel
(1.32mm per pixel)
0.066″ per pixel
(1.67mm per pixel)
-0.014″ per pixel
(-0.35mm per pixel)
25 feet (7.62m) 0.065″ per pixel
(1.65mm per pixel)
0.081″ per pixel
(0.689mm per pixel)
-0.016″ per pixel
(-0.42mm per pixel)

 

Table 5: Summary of Calculated Resolution from Actual Imagery Compared to Hypothetical Resolution, 42mm Focal Length

 

Distance 42mm Hypoth 42mm Actual Difference
5 feet (1.524m) 0.0054″ per pixel
(0.138mm per pixel)
0.006″ per pixel
(0.15mm per pixel)
-0.0006″ per pixel
(-0.012mm per pixel)
10 feet (3.048m) 0.0109″ per pixel
(0.275mm per pixel)
0.012″ per pixel
(0.294mm per pixel)
-0.003″ per pixel
(-0.019mm per pixel)
15 feet (4.572m) 0.0163″ per pixel
(0.413mm per pixel)
0.017″ per pixel
(0.427mm per pixel)
-0.007″ per pixel
(-0.014mm per pixel)
20 feet (6.096m) 0.0217″ per pixel
(0.551mm per pixel)
0.023″ per pixel
(0.588mm per pixel)
-0.0013″ per pixel
(-0.037mm per pixel)
25 feet (7.62m) 0.0271″ per pixel
(0.689mm per pixel)
0.029″ per pixel
(0.73mm per pixel)
-0.019″ per pixel
(-0.041mm per pixel)

 

Imagery from each focal length tended to have a slightly lower resolution than the hypothetical calculations suggested. However, at a 14mm focal length, the difference increased significantly with distance. I am unable to determine the cause of this growing difference, but I hypothesize that the reduction of clarity in the 14mm images from longer distances resulted in measuring error when the imagery was processed. However, resolutions that are largely quite close to the hypothetical calculations are an encouraging sign that the sensor will be able to perform very well as a high-resolution remote sensing imager.

With such impressive resolution from either focal length at five or ten feet, it will be possible with this drone to potentially map an area at a level that can identify and track sediment transport and deposition, at least with grain sizes down to 0.25mm.

So, the big question: is this the beginning of a micro-mapping revolution? Not so fast. To achieve such high resolution imagery, spatial coverage of each image is sacrificed. As Table 1 showed, decreasing distance between the sensor and the target results in far less spatial coverage.

What does this ultimately mean? The tables below display the coverage for single images at each distance and corresponding data needs to cover a real-world area unit; Table 6 displays the number of images needed at a 14mm focal length to cover one acre (43,560ft2 and one hectare (10,000m2), while Table 7 displays the same for a 42mm focal length. These numbers are hypothetical, in that they don’t account for imagery overlap, which is necessary to stitch the images together but can also unintentionally occur to a larger extent than necessary.

Table 6: Spatial Coverage Limitations, Minimum Number of Images Needed per Acre and Hectare, 14mm Focal Length

 

Distance Coverage Img/Acre Img/Hect
5 feet (1.524m) 16.67ft2
(1.548m2)
2,614 6,460
10 feet (3.048m) 66.67ft2
(6.12m2)
654 1,633
15 feet (4.572m) 150ft2
(13.935m2)
291 718
20 feet (6.096m) 266.67ft2
(24.774m2)
164 404
25 feet (7.62m) 416.67ft2
(38.71m2)
105 259

 

Table 7: Spatial Coverage Limitations, Minimum Number of Images Needed per Acre and Hectare, 42mm Focal Length

 

Distance Coverage Img/Acre Img/Hect
5 feet (1.524m) 3.26ft2
(0.302m2)
13,362 33,113
10 feet (3.048m) 13.02ft2
(1.21m2)
3,346 8,264
15 feet (4.572m) 29.3ft2
(2.72m2)
1,487 3,677
20 feet (6.096m) 52.06ft2
(4.838m2)
837 2,067
25 feet (7.62m) 81.41ft2
(7.562m2)
536 1,322

From these tables, we can see that mapping large parcels of land at the extremely high resolution provided at low distances would be largely unfeasible, because of data requirements and work-time necessary to stitch together that number of images.

While it might certainly be possible to get imagery from an acre at a 0.006″ per pixel (0.15mm per pixel) resolution by flying the drone at an altitude of five feet with a focal length of 42mm, processing, georeferencing and stitching over 13,000 images would take years to achieve. However, for small study sites of a few hundred square feet with need for extremely high resolution imagery, this could be a great solution.

At the same time, the 105 images required to map an acre at at the lowest resolution imagery from the test is a great deal more feasible, and the 0.081″ per pixel (2.07mm per pixel) resolution that arrangement offers far exceeds most remotely sensed imagery available to mappers today.

Because other variables, such as the drone’s speed, accuracy, and ability to not overlap coverage are unknown until the equipment is tested, further application of these findings will have to wait until the drone is delivered over the holiday break.

I don’t know about you, but with these capabilities, I’m looking forward to its arrival.

Work Cited

Walker, J. Douglas and Harvey A. Cohen. 2009. Geoscience Handbook: AGI Data Sheets, Fourth Edition. Alexandria, VA: American Geosciences Institute. 310 pp.

Author: Andrew Shears

Andrew Shears is an Assistant Professor of Geography at Mansfield University in Mansfield, Pennsylvania. His research interests lie at an intersection of the human-environmental nexus, and includes branches of mapping, technological, memorialization and urban geographies. He lives in Wellsboro, Pennsylvania with his wife Amy, a professional photographer.