Efovee

About

Efovee provides custom geometric camera calibration and validation services for your camera's needs. To put it simply, camera calibration allows a camera to understand an object's location in the real world, like a ruler, as seen below where the camera measures the 3D coordinates of the stop sign.

High accuracy geometric camera calibration is becoming more important in machine vision industries such as automotive, aerospace, and many others. At Efovee we strive to understand your camera's accuracy needs, and we provide meaningful feedback on what accuracies can be achieved - more than just reprojection error.

Efovee currently provides the following services to its customers:

Camera Calibration

Efovee calibrates your cameras for you. We provide custom calibration to get the highest accuracy camera model possible.

Benchmarking

Efovee tests your cameras to determine what accuracies can be achieved, to perform competitor analysis, or assess feasibility.

Validation and Analytics

Efovee validates your cameras at range providing you with a meaningful real world understanding of your camera's calibration accuracy.

Efovee is currently working on the following features...

Online API

If you already have your own calibration setup, you can call the Efovee API to calibrate your cameras.

Test Fixture

Efovee is developing a calibration machine that will automate the calibration process. Current calibration runtime is around 10-20 seconds depending on field of view.

Calibration

Efovee provides custom geometric calibration solutions to get the best accuracy model derived from your cameras. We are able to meet very tight accuracy requirements achieving accuracies orders of magnitude lower than open source alternatives and competitors.

We achieve such accuracies by customizing unique solutions for every camera module. Our calibration process and software are very flexible, so Efovee is easily tailored to your needs for a quick turnaround time. Calibration statistics are also provided which will help you understand how your camera meaningfully performs in practice.

Efovee Calibration Example

Focal Length (FLx, FLy): [782.4555721748328, 781.1069995503218]
Principal Point (PPx, PPy): [942.6003209107129, 550.7687616615057]
Center of Distortion (CoDx, CoDy): [944.872667906343, 553.0388902378644]
Distortion Coefficients (k1....kn): [-8.62715229e-04, -6.58635305e-04, -7.96401981e-04, -4.27027830e-04, -2.81477023e-06, -9.33947521e-04]

Validation

Reprojection error is certainly a meaningful statistic ubiquitous to geometric camera calibration, but reprojection error may be biased by the bundle adjustment while estimating the parameters resulting in an inaccurate calibration.

So how does the camera perform in reality?

The triangulation test shows how accurate a camera performs in the real world. This test is performed at range, and encompasses the entire field of view to ensure a calibration has uniform accuracy across its image.

The distance between error for each triangulation target pair may be seen in the error plot below. In this error plot, 10.5mm in error is equivalent to 0.15% in relative error, or 0.03 degrees in error. This level of achieved accuracy is for a $20, 100° Arduino camera.

*Lens: Arducam M40180H10 (100deg) | Sensor: Sony IMX323

OpenCV Comparison

It can be seen that Efovee's method is 100 times more accurate than open source alternatives - such as OpenCV - which is accurate on the order of meters, not millimeters.

Typical Recommended OpenCV Capture Plan

Focal Length (FLx, FLy): [779.3078823383955, 779.5024261178617]
Principal Point (PPx, PPy): [935.4601002958335, 548.1790814974687]
Center of Distortion (CoDx, CoDy): [935.4601002958335, 548.1790814974687]
Distortion Coefficients (k1....kn): [-0.1211714968834584, 0.020701251377411144, -0.016331156539148923, 0.0045673639712470945]

Contact

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';