I am working on a nature preserve project located in Indiana. The property is 249 acres and includes forests, prairie, wetlands, and river habitat. I am using a DJI Mavic 2 Pro and Pix4DCapture Pro for iOS. I divided the property into 11 sections to make each mission more manageable because I only have three batteries and the project guidelines recommend smaller missions for large projects. All flights were conducted at 50 meters with 1.17 cm/px GSD, 80% front overlap, 70% side overlap, and a -90-degree camera angle. On average each section/quadrant is 11 hectares and has approximately 600-750 images. After processing each mission separately in Pix4DMapper, my quality reports show consistent dataset, matching, and georeferencing issues. For example, mission4 shows 862 of 883 images calibrated (97%) with all images enabled and 57 blocks. I would like to reduce the number of blocks and increase the number of matches per calibrated image. I have tried and failed at adding MTPs and/or changing initial processing parameters/options. For example, I set the processing options (settings) for mission4 at [keypoints image scale = full, image scale:1] and [advanced: matching image pairs = aerial grid] and [advanced: matching strategy = geometrically verified matching: no] and [ advanced keypoint extraction = targeted number of keypoints: automatic] and [advanced: calibration = geolocation based + all prior internal + all external + rematch: no]. These settings are available in the attached quality report.
- How do I improve my quality checks for better results? Do you recommend more overlap? Example: 85% front and 85% side.
- Does this type of project require ground control points (GCPs)?
- What altitude do you recommend? Example: 70 meters instead of 50 meters.
mission4_report.pdf (1.9 MB)
Mapping heavy vegetation, like a forest, is inherently difficult. Vegetation has a very complex geometry that can make it difficult for the software to generate a sufficient amount of keypoints. Furthermore, vegetation can also move in the wind. This further exacerbates the challenges of tying everything together.
The solution is to fly with a very high overlap. Perhaps up to 90% front and side. Furthermore, you flew at 50m. This is rather low and you will definitely have better results if you fly higher.
In your current dataset, you should try processing at a lower image scale for step 1. Try dropping it to 1/2 or 1/4. This will tend to negate the effects of the vegetation and help calibrate the scene better. T
To answer your questions:
- 85% might work well. If it is heavy veg then try up to 90%. But do not exceed 90%.
- GCPs are not required to improve results in the vegetation. You can add MTPs around your project and this will also help.
- Fly as high as you can. It will help!
Thank you, Mike_K. I appreciate your taking the time to read and respond to my post. I was advised mapping forests is challenging, but didn’t realize how challenging the work actually can be. Good point - - I never thought about forest canopy movement. That makes a lot of sense. I don’t like flying with wind speeds above 12 mph. I know the Mavic 2 Pro can handle higher wind speeds, but that makes me uncomfortable as I am still a beginner. I think all my flights for this project are around 5-9 mph. Hopefully collecting imagery when wind speeds are lower helps mitigate distortion and movement as you pointed out. I also collected imagery in December without leaves as I am interested in viewing the forest floor. I don’t have access to LiDAR equipment so early winter was the best compromise. Since my last post, I contacted a former coworker who oversees the archeology lab at my university. He has a Carlson BRx7 with a base and rover. We coordinated a site visit in a few weeks to collect GCPs before I had the chance to read your reply. While awaiting your recommendations I presumed GCPs could help overall quality.
- Will adding GCPs help or hinder the quality of my project?
- Can I still add MTPs after GCPs?
Hi Mike K,
On 2/15/24 I am returning to my project site to place GCP photogrammetric targets on the forest floor and record their locations with RTK equipment. The forest site is 249 acres with an elevation change of 33ft (north/south) and 24ft (east/west). The elevation is nearly flat and has no visible obstructions except trees. The site has one creek dividing the property north/south and one river dividing the property east/west. I am attaching some visuals for the property. I know P4D requires a minimum of three GCPs and explains 5-10 are generally enough for large projects. However, the guidelines do not define “large” projects so I do not know if 249 is considered small, large, or somewhere in between. I selected locations for 11 GCPs and marked them in the attached visual. Thank you for all your help.
- Are 11 GCPs too many for the size of my project?
- Is the spacing and distribution of the marked GCP locations sufficient?
- Are there any factors I am not considering?
- Do you have any suggestions or recommendations?
I took a look at the images that you showed, and they look good to me. Having 11 GCPs is not going to be too many, especially given the topography of the project. Here at Pix4D, we recommend 5-10 since after that, the amount of precision gained diminishes, but the ideal number depends on the topography, complexity, and size of the project area. I would say that your distribution looks quite good, especially if the terrain is relatively flat. Since it is heavily forested, it will be quite a complex project to do. Just be sure that where you place the GCP targets can be seen clearly in as many images as possible without being obscured by trees or shadows. When marking the targets, zoom in as far as possible to mark the center on a pixel level.
When recording the GCPs, you might have problems with a good signal depending on the canopy and the quality of the satellite connection. Be sure to place the targets in clearings as much as possible, otherwise your measurements might not be very precise. Using clearings will also help with the visibility of the targets. The targets themselves should be ten times the GSD of the project (if you have a GSD of 1 inch, your targets should be at least 10 inches x 10 inches).
Given the size of the project, if you have to fly multiple missions, be sure that there is a significant overlap between missions so that the software can combine them more easily.
As Mike said, make sure that you fly relatively high with a high overlap to ensure that the vegetation is not moving too much between the images. Try to fly at least one flight line beyond the boundaries of the project to ensure good coverage of the edges of the project area.
Good luck and enjoy your day in the forest!
Dr. Ryan Hughes
Pix4D Senior Technical Trainer and Content Creator
Hi Dr. Hughes,
I want to ensure that my targets are big enough based on your recent feedback. The example you mentioned explains that “targets themselves should be ten times the GSD of the project (if you have a GSD of 1 inch, your targets should be at least 10 inches x 10 inches). I attempted to use the P4D GSD calculator and found based on my inputs that I have a 1.33 cm GSD. That seems very small so I want to double-check with someone more experienced that I used the correct inputs. I am including a screenshot of the P4D calculator inputs and results. I am using a DJI Mavic 2 Pro for my project. The specifications for my UAV show a 25.4 mm or 1” sensor width, 28 mm focal length, 5472 imW, and 3648 imH. I plan to fly between 80 and 90 meters. Buying targets was too costly, so I made targets for GCPs using inexpensive high-contrast tent fabric cut into 12x12-inch squares, which I assembled into 2x2-ft squares with a sewing machine.
- Did I enter the correct inputs into the calculator?
- Are the dimensions of my targets adequate?
- Is there any disadvantage when using large targets?
Theoretically, I would say that those targets would work just fine in terms of size. The best way would be to test in the field.
Take your drone to the desired flight height and take six images in a small grid pattern over the target. Then, upload those images to PIX4Dmapper/PIX4Dmatic to be sure that you can mark the center as precisely as possible. You can repeat this procedure with different flight heights until you reach the upper limit and you can no longer clearly identify the center of your target (always stay within the legal flight height…).
This is the procedure I use personally when testing new targets. This way, I know exactly the maximum flight height for each of the targets, and I don’t have a problem in the field. What I have done in the past is to mark on my carrying case the maximum flight heights for each of my targets with the megapixels of the camera used (for example, 60 meters x 12 MP). This way, I don’t forget when I haven’t used those targets for a long time.
There are only slight disadvantages to using large targets. The first is transport, but if yours are fabric, this is less of an issue. In terms of placement, you will need to ensure that the target is as flat as possible. This can be an issue with fabric targets depending on the terrain. You can always place cardboard or another solid surface underneath so that it sits flat (for example on grass or low vegetation). In the image that you show, the target seems fine to me. The final disadvantage that comes to mind depends on the size of the project area. If you are doing a small project area, larger than necessary targets look less nice. But it is primarily a question of aesthetics.
Thank you, Dr. Hughes. Your guidance about sizing targets by conducting some test flights and recording the height and megapixels makes sense. I still have uncertainty about the GSD Calculator and the inputs for my DJI Mavic 2 Pro. Should I use the GSD calculator? Are my inputs correct? And if so, what does 1.33 centimeters/pixels mean? Lastly, how would I use the GSD calculator answer (e.g., 1.33 cm/px) to determine the size of my target related to the size of my project? I am m not sure I fully understand what I should be doing with the results from the calculator after I have my number.