This project consisted of several phases and took place in an office building of Mosini Caviezel SA, Switzerland: a rural house that includes five rooms and a large hallway.
The main objectives were to create an indoor plan in 2D, evaluate the application of terrestrial photogrammetry for establishing a PPE (Propriété Par Étages/property by floors, or condominium ownership of an apartment) plan, and find out an optimal protocol for surveying and architecture companies to document property indoors that is of low cost and sustainable.
The first phase was to acquire images in different ways, to figure out the ones which give the best results. Using the photogrammetry software Pix4Dmapper, it was easy to assess the quality of each method and to continue to the following step – scaling the project to obtain precise measurements.
In the final phase, all processed Pix4D projects were analysed and compared in order to assess the feasibility of the workflow and to sum up for a protocol in the near future.
All images were taken, following some camera settings below, with a GarminVIRB ™ Elite:
– Fixed focal length or disabled auto-focus for stable camera interiors
– Large and fixed shutter in order to receive more light
– Low ISO values to reduce noise and improve image quality
The sports camera reduced motion-blur and the wide angle reduced the need for as many images, as it covers wider views. This is especially important for a narrow space indoors. After picking the preferred camera, Mosini Caviezel SA needed to know the optimal image taking plans that would give the best results. In this step, they took images in three different positioning ways.
Test 1: Stand in the middle of the room and turn around while taking images
Images were taken perpendicular to the measured area. Baseline equaling to zero created sharp-angle intersections which led to unsatisfactory results.
Test 2: Across the room along the long side, positioned in middle of the two sides
It is generally difficult to take images crossing a narrow space, and thus this type of configuration was only applied to corridor zones. Area of interest is always at the edge of images where more distortion occurs.
Test 3: Move parallel to the walls
In this configuration, the operator will move around the room and always face perpendicular to the area being measured. Area of interest appears at the center and thus there is less distortion in the images. Among the three walking ways tested, this one is the most favoured one.
After picking out the best configuration of walking around the measured space, the following step is to find the optimal camera shooting angles. Here Mosini Caviezel SA compared the actual matching results from various acquisition angles in different survey areas.
Method 1: Back to the wall, shooting at 90 degrees
With this method, 126 images were taken and analysed. Using this method, the matches in Pix4Dmapper were the most numerous and well-distributed. Moreover, there was less reprojection error (0.2708 pixels) than all the other methods.
Median points of interest per image: 21709
Median of connection points by pairs of images: 2550
Method 2: In the middle of the room, shooting straight forward
81 images acquired with this method were processed. The sharp intersection angles do not generate good matches. This reason of this unsatisfied result was similar to test 2 above. Pixels are obviously distorted on the edges, which leads to less matches among images than method 1, though similar projection errors (0.2655 pixels.)
Median points of interest per image: 21455
Median of connection points by pairs of images: 1756
Method 3: Back to the wall, shooting at 45 degrees
Only 58 images were needed using this method. The method forms the intersection angles slightly larger (better in this sense) than method 2 but less image overlap was obtained. No significant difference in projection errors (0.2469 pixels) was found, but fewer matches were observed. Matches were only found on the ground, and there were nearly no matches on the walls, the area of interest.
Median points of interest per image: 21075
Median of connection points by pairs of images: 1578
Method 4: Back to the wall, shooting at 60 degrees
64 images were taken with this method. It produced slightly better distributed matches compared to method 3, though still concentrated more on the ground. The reprojection error is 0.2266 pixels and the number of total matches are similar to the other methods.
Median points of interest per image: 21080
Median of connection points by pairs of images: 1492
From the above data acquisition methods, the first method was the most favored one, generating the best matches among images. The only disadvantage of this method is the narrower coverage and the impact from close-object obstructions. More images would be needed using the method 1 acquisition orientation.
In order to obtain precise measurement of the reconstructed result, it is very important to assign correct scales under non-geolocated situations. Mosini Caviezel SA placed targets on the walls, making sure the scales were given in two perpendicular directions to ensure the scale is correct in all directions. Four control points, with local coordinates, were placed around the room, and five check points were assessed as follows.
Accuracy of the entire reconstruction is not as consistent as outdoor cases. However, by assigning the correct scale to the project, general measurements could get more precise.
Advantage of This New Technology
Indoor mapping has been one of the challenging area for photogrammetry. However, it can be an efficient and low-cost way to obtain a precise 2D floor plan using images and Pix4Dmapper Pro.
– Ideal method for reconstructing areas with complicated geometry
– Intuitive and easy-to-use interface, small learning curve
– Affordable license for new equipment investment
– Scalable platform and high-performance customer service
Author: Valérie Luthi
Suggestions from Pix4D
Due to dim light conditions and narrow space, applying photogrammetry indoors is more challenging than outdoors.
Here we proposed several solutions for people who would be interested in giving this a try.
If the purpose is to obtain a 2D plan, a device with 4K video would give you more continuous results. Pix4Dmapper Pro processes videos and extracts them into still images, and by using the video as input, you can ensure high image overlap. This is the most convenient way for acquiring a simple 2D plan, but the low resolution and large distortion will not give you a nice 3D reconstruction.
- Fisheye vs. Perspective Cameras
Perspective cameras will give you better 3D reconstruction output. However, you will need to have some experience with how to acquire images with sufficient overlap in closed area. Fisheye cameras or spherical cameras will make your image acquisition procedure a lot easier, but you will expect noise on the edge induced by the large distortions. If you look for survey-grade accuracy, perspective lenses are still the way though more advanced knowledge and training will be needed.
For users looking for something in between, spherical camera is a good option. Here you may download our sample dataset to experience your first indoor mapping with Pix4Dmapper Pro now!