Abstract

In order to rehabilitate, repair or remodel buildings it is necessary to know the position of the installations in the wall, such as pipes, cables and structural elements. Currently, this is still determined by blueprints or special sensors. With AR it is possible to project the installations on the wall via smartphone in order to accurately represent the course of the pipes, cables etc.

The idea is now to design a so-called frame marker that has a cutout in the middle, so you can attach it to light switches and sockets. The advantages of that, compared to already existing markers, are that the frame marker on a socket or light switch is much more unobtrusive and does not need to be extra marked in the blueprint since it has the same position in the construction plan as the socket or light switch.

Furthermore, the marker is tested in terms of the quality of detection, the maximum detection distance, and the viewing angle.

Implementation

The general approach of the implementation is first to take the RGB picture taken by the smartphone's camera and convert it into a grayscale picture. After that, we analyze the picture for edges and sort those contours out, which describe a quadrangle. These contours are marked as possible frame marker candidates. Now for each candidate, we analyzed the encoding and performed some error correction on it. If the code matches the code we are looking for, that means that we have found a frame marker. After detecting the marker, we perform pose estimation in order to get the position of the marker in the room.

Conclusion

The results were that paper marker was still recognizable at about a distance of 120 cm, whereas the maximum detection distance of the socket frame marker was about 80 cm. The maximum possible viewing angle of the socket frame marker was also significantly worse (about 45°) compared to the paper marker (about 60°) at about a distance of 50 cm. The reason for this performance difference is the rounding of the edges of the socket frame. The  socket frame marker could only be recognized at a distance less than a meter. Therefore it is not possible to have a view of the whole wall, where the marker is placed, because the marker will be out of reach. A solution for that would be combining the current detection system with a marker-less tracking technique like e.g., ARCore