When measuring 3D Depth Cameras quality and performance, establishing a common attribute is best to characterize the measurements to represent its performance, hence the preferred metrics but not limited to are its Z-Accuracy, Precision, setting a reasonable Region-of-Interest, and a high-fidelity Point Cloud.
Z-Accuracy #
Z-Accuracy measures how close the depth values are close to the actual distance (the Ground Truth). The Z-Accuracy is represented in percentage (%), meaning its variation on its closeness of a measured value to a true value. When looking at an accuracy chart, it is important to consider that a chart is representation of the depth readings at each distance. As the % defines the variation, the closer to zero in its percentage value, the higher the accuracy it is since the variation is minimal closing to zero.
Upon different techniques, improvement and enhancement are made varying each 3D camera to get its best performance for which the cameras are intended for.
In the next chart we can see comparing our LIPSedge™ AE400 vs. LIPSedge™ AE430, the Depth performance was quite improved to meet industry demands.
Comparing the Accuracy between LIPSedge™ AE400 and LIPSedge™ AE430:
The LIPSedge™ AE400 as meant for industrial application use-case had an accuracy from a 0.4% to 1% measured from a range of 1 meter to 3.5 meters, on each distances respectively.
The LIPSedge™ AE430 a light-weight industrial 3D depth camera for more specific uses such as Robotics, required a more accurate readings, hence adjustments were made to improve its depth performance.
Precision #
As for Precision, represented by using a metrics value, e.g. millimeters, is the actual distance refers to the closeness among successive measurements carried out with the same method by the same operator under identical conditions.
When reading a precision chart, since the values are represented by the iterations of measured values, the lesser the distance the better the precision is, as it would mean the measured values are not scattered, a direct representation of an optimal condition from the optical calibration.
In the next chart, it can be observed for the short-range measurements from the 40 cm to 70 cm range, although the variation compared between the LIPSedge™ AE430 and a Legacy Stereo Camera stand close, the minimal difference is actually a huge gap for what precision demands when detecting at close distance, demonstrating the importance and advantage about readout and the results obtained from the LIPSedge™ AE430.
Region of Interest #
Region-of-Interest (known as ROI) is referred to a predefined area from within the Field-of-View of the camera. Such area is adjusted by the user depending on the target object to cover. The ROI would vary upon where the camera is located towards the distance of the target object. A user would need to adjust it based on the desired coverage. An example is if a coverage of a target is satisfactory at 80% of ROI within at 1 meter, if extending such distance to 2 meters, then such coverage would be only 40%.
Point-Cloud #
A composition of all factors from the 3D Camera’s specifications, including:
- Accuracy
- Precision
- Noise
- Ambient conditions
for which would affect depth performance and its results.