We were invited by a local council to demonstrate our aerial topographic survey capabilities for a small culvert project. This was challenging as the route passed through particularly dense tree coverage. Typically multi return LiDAR would offer better penetration to identify true surface though vegetation and is typically chosen in this situation (See our article comparing Photogrammetry and LiDAR). In this instance we wanted to show photogrammetry as a valid option during winter months which would have a much smaller cost to the client over LiDAR data collection (typically costing thousands a day). The survey manager was specifically interested in a topographic drone survey to see how this compared with existing measurements of the site. The culvert ran next to an existing tarmac walkway under some small tree’s densely packed together.
In February we returned to site to conduct the survey. This was to be a geo-referenced survey and we chose to use ground control points and manually collect their locations using a survey pole and base station for later post-processing (PPK).
The base station was firstly set up to be run for 2 hours. In this case we used the Emlid Reach RS GNSS that we use for our own RTK drone setup. However, we are happy to use which ever survey equipment clients prefer such as those from Leica and Trimble. In our test we knew the accuracy of the Emlid would still be several magnitudes better than the accuracy of positioning information available from the drone. We laid out 12 ground control points along the length of the route with square/diamond patterns indicating the center point where the position would be taken with the rover pole.
In this case since the mission was a demonstration and we were close to residential property we chose to use the DJI Mavic 2 Pro. Typically we would use the DJI Matrice, S1000 or heavy lift drone with a full frame camera such as the Sony A7 range. For mission planning we chose Drone Deploy with a one pass Nadir grid with standard settings at a low altitude of 20m. The mission was flow collecting 199 images of 5472×3648. Some blurring was observed due to poor winter light. After the mission was flow we measured the position of the GCPs before clearing up.
Survey Data Processing
Later the survey data was post processed using RTKLib and local RiNEX data to provide accurate positions of each GCPs. This provided an accuracy under 2cm X,Y and under 4cm in height. The images and GCP locations were input into Pix4D and the mission processed to obtain a classified point cloud, DEM, DTM, mesh model and contour map as show below. This had an ground sampling accuracy of 0.81cm/px over the area of 2.47ha. The absolute camera position uncertainties X,Y,Z(sigma) were 0.112(0.021), 0.317(0.028) ,0.71(0.051)m. The error of a couple of the GCP measurements was high as the rover had poor line of sight to the base station on the hill through vegetation.
From the results we observed holes in the point cloud due to the dense wood branches over the pathway. This led to a poor contour model as seen below. This was disappointing but highlighted the importance of planning in accordance with the task which Drone Deploy alone didn’t consider.
A Second Try
We returned and manually flew and acquired photos around the problematic area this time taking additional oblique images perpendicular to the walkway. Processing these images provided an improved point cloud with the walk way visible through the trees to measure the topography along the route.
The benefit from the extra processing to classify the point cloud was that high vegetation was classified and could be removed from the point cloud to provide a more accurate DSM for measurement and topographical contour creation.
This example from the earlier batch of processing without point cloud classification and high vegetation removal shows peaks along the DSM profile.
Our conclusion highlights that during winter months photogrammetry can provide actionable topographic survey data over woodlands, an area where LiDAR is typically required. With anticipated or non essential delivery schedules this can provide huge financial savings. Sky Tech are happy to work with local surveyors on projects such as this where drones can provide more rapid data collection over convential site walks. Please get in touch if you would like to work with us:
This article covers the state of the art in drone LiDAR and photogrammetry to capture 3D information for survey and construction. The advantages of each method will be covered in terms of accuracy, complexity and cost. Using drones for survey offers huge time savings over manual measurement on the ground. The data recorded can be processed and used for asset classification, validation as well as health and safety. The ease of use means that regular scanning can take place providing a time series as work progresses. This can be used to monitor progress, detect issues as they arise, improve management and reduce risk.
Drones For Surveying
Construction drones can assist in planning stages providing anything from aerial photographs, video to more complex modelling using LiDAR. Aerial photography of a site in the landscape is useful for architects and planners to conceptualise how the new development will look in it’s end setting. This is also possible with sophisticated video tools. An easier approach is to use photogrammetry and/or LiDAR to create a 3D model (mesh) of the larger area. Within architectural software the new building would be added. From here virtual fly throughs can be created of how the completed site will look. The point clouds derived from Photogrammetry and LiDAR mapping can be used to build a digital terrain model (DTM) or digital surface model (DSM). This forms part of creating a topographical survey of the area for flood assessment, scrap and fill operations, laying out the site and planning foundations along with wider geotechnical subsurface survey.
Drones For Construction
The constuction industry is now widely
adopting building information modeling (BIM). A concept within this
is the digital twin where every element of a building has a digital
representation down to the last screw and fitting. From a
construction point of view BIM provides less friction between trades
from quoting to assembly as each step of design and construction is
laid out ahead of time.
Aside from survey operations, during
the construction phase drones can be used to provide an overview of
the site to monitor progress, foundations, build, stockpiles, bottle
necks as equipment/stock is moved around and overall safety. As
construction progresses drone imagery can be used to create 3D models
with millimeter measurement accuracy to verify what has been built at
each stage matches the digital model.
What Is LiDAR?
Light Detection And Ranging (LiDAR) is a method of measurement using a pulsed laser and sensor. As a spot on the ground is illuminated the difference in reflected light return time and wavelength is measured. A LiDAR scanner system typical fires thousands of pulses per second. The raw data is typically processed and georeferenced into a 3D visualisation know as a point cloud. These sensors are split into two categories: spinning prism with 360 degree coverage or phased array with a narrower field of view.
What Is Photogrammetry?
Photogrammetry is a method to obtain, record and measure a series of images to produce three dimensional information. Photogrammetry uses projective geometry theory and camera parameters such as angle, bearing, focal length, sensor size etc. All this information is cleverly combined to create 3D information. One aspect that closely flows LiDAR is the use of key matching points (the same point identified on separate images to create a point cloud.
LiDAR and Photogrammetry Coverage
Before LiDAR and Photogrammetry workflow is covered, firstly the common area of mission planning is discussed. To create a topographical map or 3D model it is necessary to scan in a predetermined grid or pattern that allows sufficient coverage and overlap of the area. DJI have made it easy for other companies to control their drones by releasing a developer kit. Popular apps include DJI Mission Planner, Drone Deploy, Hammer and Pix4D Capture. The open source equivalent PX4 used in custom drones also has mission planning software from UGCS, Auterion, Ardupilot and QGroundControl.
Velodyne has dominated the LiDAR sensor market and appears in several drone LiDAR solutions from Route Scene, Yellow Scan and Phoenix UAV among others. These typically have an accuracy of +/- 2cm at 80-100m range. Along side the sensor these drone LiDAR solutions make the addition of a very sensitive IMU and RTK sensor for global positioning. The combination of all the system components is considerable, requiring a heavy lift drone platform with a relatively short flight time over using just a camera. The best systems also include a full frame DSLR camera as well.
Drone LiDAR requires scanning with significant enough overlap to capture common scan returns. For mapping LiDAR would be captured in a grid pattern from a set height. For modelling a building LiDAR would be captured in an orbit at different heights.
After flight further data processing is required, combining multiple laser returns, IMU position and satellite position to create a precise geo located point cloud. This is computationally demanding to align multiple returns into the final point cloud. The chosen output is closely tied to the selected industry be it construction or land survey. In construction software such as Autodesk and Solidworks are used. These have a growing role in BIM for design validation during construction using interior and exterior LiDAR scanning. In land survey QGIS, Global Mapper and QGIS now have LiDAR modules.
Photogrammetry with drones requires
certain camera parameters and capture methods. Camera metadata
parameters must be know such as sensor size, pixel count, focal
length, shutter type, tilt, bearing, roll and latitude and longitude.
This information is contained in photographs taken from most off the
shelf drones. In special cases external large format cameras with
global shutters and very high pixel count are desirable such as the
Sony A7 or Phase One.
Photogrammetry requires images to be
captured in sequence and with significant enough overlap to capture
common keypoints. For mapping images would be captured in a grid
pattern from a set height. For modelling a building images would be
captured in an orbit at different heights.
In turn for their small size, low cost
and ease of use, DJI drones cameras have lower quality images than
full frame cameras. Drones such as the DJI Phantom 4 Pro and DJI
Inspire 2 can still provide excellent results as long as lighting is
correct at the time of survey. The accuracy of photogrammetry
relative to the sensor size and height and known as Ground Sampling
Distance (GSD). This is proportional to the height the drone is
flown. Higher accuracy at lower altitude is a trade off in longer
flight times to cover a tighter grid to maintain overlap.
Photogrammetry processing is more computationally demanding than drone LiDAR requiring a high spec PC with additional GPU power. Major software distributers include Agisoft Metashape, Pix4D and Reality Capture. Each has it’s merits for survey or 3D modelling. Some even combine LiDAR and photogrammetry to improve results.
Does Drone LiDAR Provide More Accurate Measurement Than Photogrammetry?
It is difficult to discuss the merits of each as there are several ways to obtain accurate position. Ground Control Points are widely used in aerial survey but are more tangible for Photogrammetry to use. These are simply artificial boards 1m squared or larger with a target painted on. Survey equipment is used to measure the center point and hence tie the recorded digital map back to set geological coordinates.
Drone LiDAR and Photogrammetry positioning relies firstly on GPS. This provides accuracy down to 5m enough for flight planning. Exact geotagging requires sub centimetre accuracy during flight and more satellite information is needed. Real Time Kinetic (RTK) is one method using a drone base station to transmit satellite corrections to the drone over a correction network. Post Processed Kinematic (PPK) is another alternative where network access in the wilds isn’t always possible. This takes all the information from the drone, base station and corrects it using a local correction station.
The accuracy of drone LiDAR and Photogrammetry using the positioning methods above should be comparable in favorable circumstances. LiDAR is limited by the spot size of the laser, Photogrammetry by the number of key point matches. For smooth areas such as water, snow or concrete photogrammetry will loose resolution.
Does LiDAR Provide Better Quality Results Than Photogrammetry
LiDAR sensors benefit from having multiple returns allowing the laser to pass through light foliage to measure true surface level. Photogrammetry in this case has less resolution returning only the terrain height from hedges, trees and grass. Software can help this issue to an extent by classifying the point cloud and removing the higher areas classified as foliage to get closer to the true surface level.
An immediate draw back of LiDAR is that true colour is not captured and a limited colour pallet is available. This limits it use alone when creating photorealistic aerial maps and 3D models. LiDAR has problems with very black surfaces which absorb the laser light or reflective surfaces which scatter it.
Comparison of LiDAR and Photogrammetry
£200-300K for drone, ground station and dedicated software.
£2-100K drone from low quality Mavic 2 to heavy lift with gigapixel camera.
Few specialist operators.
Lots of operators to choose from.
Specialist software to compute georeferenced point cloud.
Lots of options, dedicated PC required.
1-2cm vertical, 2cm horizontal at 100m.
2-3cm vertical, 1cm horizontal depending on camera resolution.
Laser technology understood by industry. Vegetation less of a problem.
Lots of drone options. Easy to use software.
Expensive technology with limited access thats hard to use. Limited colour palette.
Requires good light, poor in shadows of smooth areas. Accuracy poor without RTK or GCPs.
3D point clouds with return information suitable of classification.
2D maps, 3D models, point clouds surface models with visual detail.
Terrain modelling, 3D modelling of complex structures, steel, pipes.
The best solution is to combine both LiDAR and Photogrammetry methods to compensate for the short comings. Each has it’s merits depending on use, budget and schedule. It comes down to understanding these and knowing in what application each method has it’s strength and weakness. This will be covered in a series of upcoming case studies. The cost of LiDAR will continue to drop in the near future. Processing will also become more and more sophisticated with increasingly transparent to understand deliverables.
In construction for BIM, asset classification and tracking Sky TechLimited can provide accurate survey and 3D models using drones to meet you needs, contact us today: