The comparison between LiDAR and photogrammetry has recently become a subject of continuous discussion in the aerial surveying community, since they are the two most popular methods in the field. This was also what led us to perform a test survey using these two methods in the same project area to derive the same product – an accurate digital terrain model (DTM) of a breakwater.
For the test area, we chose a breakwater in the southern part of Scheveningen in the Netherlands. For the photogrammetric survey, an Altura Zenith ATX8 drone was used in combination with a Sony Alpha ILCE-7R full-frame camera (using 28mm focal length). Using a flying height of 50m above ground level (AGL) and a forward and side overlap of 80%, three flights were performed during which 585 images with a ground sampling distance (GSD) of 9mm were captured. For that survey, 26 ground control points (GCPs) and 16 check points were measured using real-time kinematic (RTK) GNSS and later validated with more traditional levelling. GCPs are used for the reconstruction of the model during the processing, while check points serve only validation purposes.
For the LiDAR survey, the same drone was used, this time in combination with the aerial LiDAR system YellowScan Surveyor. This system consists of the Velodyne VLP-16 (Puck) laser scanner and the Trimble APX-15 UAV GNSS-Inertial system by Applanix. During the LiDAR survey, three flights were performed, covering almost the same area (the breakwater) but with different flight plans each time. The flying height was common for all three of them and equal to 30m AGL. This height was chosen to be lower than the photogrammetric flightpath to increase the point density, since the shape of the survey area allowed us to do so, being narrow and long. For the georeferencing of the derived point cloud using RTK, a GNSS base station was set up in the project area.
The processing of the photogrammetric dataset resulted in a DTM and an orthophoto. The point density of the photogrammetric point cloud was approximately 3,500 points/m2. Figure 1 shows a closer look at a small part of the DTM. It depicts the high level of detail contained in this model with a grid size of 2cm. The DTM is characterised by smooth and clean surfaces, clear breaklines and no data gaps. In addition, Figure 2 shows a similar view of the orthophoto, with a resolution of 1cm. With such a resolution, even the smallest features can be distinguished.
To validate the correctness of the products, a quality control was performed based on the DTM and the measured GCPs and check points. To have a fair comparison, a DTM of 5cm grid size was used, instead of the initial 2cm, since the grid size of the DTM derived from the LiDAR dataset was also 5cm. That resulted in an absolute vertical accuracy (with respect to the ground truth data) of 4mm. The magnitude of the absolute accuracy depends highly on the quality of the measurements of the check point. Since the check points were measured using levelling with an accuracy of 1-2mm, the absolute accuracy that we could achieve was significantly higher than by using RTK GNSS for determining the locations of these points.
The processing of the LiDAR dataset was considerably quicker (approximately four hours versus eight hours for the photogrammetric dataset). The main product was the DTM of the de-noised point cloud. For this test, the two most precise of the three available point clouds were merged into a single point cloud to increase the point density. An average point density of 625 points/m2 was then achieved and consequently, a DTM with a grid size of 5cm could be generated.
Figure 3 is a close look at a small part of this DTM. You can see that the surfaces in this DTM are less smooth than in the DTM derived from photogrammetry. Using a different interpolation method, the result can be smoother but then we would deviate more from the true values and thus decrease the accuracy. In addition, there are some data gaps due to the presence of water.
Using the GCPs and the check points measured during the photogrammetric survey in combination with the DTM, a quality control was performed, and the absolute vertical accuracy of the LiDAR dataset was found to be 20mm. For this quality control, the GCPs of the photogrammetric survey were also used, as check points.
Testing UAV LiDAR and photogrammetry by performing the same survey on the breakwater in Scheveningen and processing their datasets led us to the following conclusions:
At first, the processing time needed for the DTM generation was significantly lower for the LiDAR dataset than for the photogrammetric one.
With photogrammetry, we achieved higher DTM resolution – 2cm versus the 5cm that was achieved with LiDAR. A higher level of detail is thus provided, something that can be very useful in certain applications that would require that.
The DTM derived from the LiDAR dataset was less smooth when compared to the one derived from photogrammetry. That can be explained by the fact that, although the LiDAR point cloud was quite dense, the point distribution was not homogeneous, since it depends a lot on the incidence angles of the scans.
Photogrammetry provided us with a very high vertical accuracy (4mm) thanks to the high overlap in the images and the very low uncertainty in the GCPs’ and check points’ measurements.
The vertical accuracy of LiDAR was also quite high (20mm), considering the specifications of the LiDAR system used – the nominal accuracy is 5cm – and is very close to the upper limit of the accuracy that can be achieved with LiDAR.
However, neither of the two methods can be regarded as better than the other. In a built environment like our test location, photogrammetry is preferable for the reasons mentioned above. But in a vegetated area or for surveying narrow structures, LiDAR is undoubtedly a more suitable solution. Therefore, the choice of the method should be made each time based on the nature of the project area and requirements.
Chara Chatzikyriakou is a project manager at Skeye BV and Patrick Rickerby is technical director at Skeye Ltd (www.skeye.eu)