This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Linux/TDA2: 3D surround view Camera Calibration tool

Part Number: TDA2

Tool/software: Linux

Hi,

I am using TDA2X vision EVM board to develop 3D surround view function. We run the calibration using the calibration tool. We found front and rear cameras can not be calibrated perfectly.

Look at the picture below. The straight line becomes curved when it passes the front camera. I think it is because the lens distortion is not completely corrected.  In the lens LUT generation tool, the maximum angles are limited to below 90. Will this be a problem? 

In another thread, you said you can provide the calibration tool source code. Can you provide that to me so we can debug what is wrong? Thanks.

1. front camera can not be calibrated perfectly. 

  • Another strange thing is that the birdview image generated by calibration tool is good. But the online birdview image has distortion.

    This picture is generated by calibration tool. The numbers in red circle has no distortion.

    This picture is captured online, the numbers in red circle has large distortion

  • Those above pictures are captured in 3d adaptive surround view function (has bowl view). I also captured some images in normal 3D surround view. There are also distortions in front and rear camera area. But the curve direction is opposite to that in 3D adaptive view. Following is the picture. Notice the curve direction in red circle. Theoretically, bird view of normal surround view should have no distortion. Why it still has distortion? Thanks.

  • Hi Tao,

    For access to SRV calibration source code, please ask the respective region TI Field representative to raise a request with me. We can pass it through the approval system and provide it to you. (I can provide the source code in public forum). This would be the source code of the embedded calibration tool.

    Regarding your questions

    • Limiting the FOV to 90degrees is not an issue, this is the radial angle, so the FOV of the lens we would capture would be 180 degrees
    • Its hard for me to debug this just looking at pictures, and I'm not sure why the calibration works fine on the offline tool but not on the device. It maybe related to the bowl shape, please make sure the bowl shape is as flat by extending the offsets to the far regions (Typically around +/-400 or so) and keep the adaptive logic turned off.Can you also mention which two usecases you tried on the device and more information on your setup (Which sensor/camera etc)

    Regards

    Shashank 

  • Hi Shashank ,

    I have figured out the cause of  issues listed above. It is because the limitation of the max angle. After I remove that limitation. Those issue disappears. But the stitched image is still not so good. Please look at the image attached below. I am using the cameras from spectrumdigital in this link: www.spectrumdigital.com/.../. The cameras don't have lenses. We bought the that and installed by ourselves. According to your experience, will those cameras be a problem?

    Thanks.

  • Hi Tao,
    Typically you need machine aligned camera modules to get good calibration. Hand screwed lenses may have issues but you may be able to get better calibration than this. You can try running the lens center detection tool from the manual calibration tool prior to generating the LENS.bin. In this case, the Lens.bin will encode the centers (I am assuming the image center and lens center do not align in this case). Modifying the lens center with the tool might help.

    Regards
    Shashank