Dragonfly is a visual SLAM technology that uses computer vision to provide the precise location of vehicles, robots, drones, forklits and moving assets. Dragonfly provides a high level of accuracy. For the vSLAM technology the accuracy can be measured in 2 ways:
1 – Drift
The “drift” is the accumulated error. Typically the drift applies only for stereo-camera installations. The drift applies only to “unknown” environments, that means venues and spaces where Dragonfly has not been used before, and for which there is not yet an existing 3D map created by Dragonfly. For example, the first time you drive a forklift with a camera and Dragonfly installed, inside a warehouse, you will be exploring an unknown environment.
The drift is expressed as a percentage: this number indicates the amount of accumulated error, and can be considered as the amount of meters of error for each 100 linear meters.
- Stereo-camera installations – Dragonfly’s drift has been verified to range between 0.6% and 1.3%. This means that if you drive a forklift with Dragonfly installed on board along a 100 meters linear path, at the end Dragonfly will report a location that can be 60 cm to 130 cm inaccurate.
- Monocular installations – the drift in itself can be pretty high but is usually easily corrected with loop-closures (see below). This is why we recommend the usage of monocular cameras only when a pre-mapping is possible. When using monocular camera the measure of accuracy is the radius of confidence (ROC). After the pre-mapping described above, the ROC is usually 1% of the distance to the closest real-world reference (marker). For instance, in a 15,000 sqm warehouse, the average ROC is about 30 cm.
The drift, and the error for monocular cameras’ systems, are however automatically corrected by Dragonfly each time there is a loop-closing. This means that each time Dragonfly recognizes an area that has already been mapped, the loop closes, and Dragonfly corrects the location and the map is also updated. Loop-closings are extremely useful to improve the overall accuracy of the system, and it is strongly recommended to perform frequent loop-closings during the initial mapping, for both monocular and stereo cameras architecture.
2 – Linear accuracy
When navigating inside a known environment, therefore when Dragonfly re-locates the device inside a previously created map, the accuracy depends on the precision of the triangulation of known points (features). The radius of confidence, is typically 5-10 cm, and depends on several factors, including:
- The quality of the camera;
- The lightning of the environment;
- The real-world reference;
- The dimension of the map;
- The camera’s calibration.
As an example, imagine a drone navigating inside a hangar that has already been “mapped” before: in this case, the ROC of the drone will be 5-10 cm.