| 40 | * We believed that the poor calibration was a result of only using a set of points that were very close to each other. |
| 41 | * Although it seemed that the absolute error was very low, this was because the points were relatively close together. |
| 42 | * In reality, the relative error was very high, and this was visible when viewing the point clouds. |
| 43 | * Our solution to this was to add more ArUco markers. |
| 44 | * We also improved our ArUco marker detection script to cache the previously detected ArcUo markers, since different markers were being detected each frame. This also improved our calibration. |
| 45 | [[Image(better_marker_detect_AdobeExpress.gif)]] |
| 46 | [[Image(12goodcalibration_AdobeExpress (1).gif)]] |
| 47 | [[Image(12calibrationtoggle_AdobeExpress.gif)]] |
| 48 | * This gif also highlights one of the main benefits of using multiple cameras. Behind the cube on one view there is an empty section that we have no information about. By combining the camera views, we now have information about this previous blind spot. |