TDA4VM: Distortion processing

I can modify the size of the output image, as shown in the following figure.

But this modification method can only cut a portion of the image from the right and bottom, leaving only the upper left corner(Rectangular area).

The existing problem is that on the one hand, it is impossible to correct the arc distortion of the red line above, and on the other hand, it will greatly reduce the camera's field of view.

Is there any other better way to solve the problem of arc distortion in the upper and lower red lines.

Regards,

Yueqian

Hi Yueqian,

qian yue said:

Is there any other better way to solve the problem of arc distortion in the upper and lower red lines.

The red area is out of your input image border.
LDC repeats the border pixels in that case.

qian yue said:

But this modification method can only cut a portion of the image from the right and bottom, leaving only the upper left corner(Rectangular area).

For rectilinear output view, it is difficult to keep the entire view without getting out of input image border, because the available view is not rectangular.

You will have to decide what you want to do first (i.e., your desirable output view).

BTW, there is a similar issue in the thread below for your reference.

e2e.ti.com/.../tda4ah-q1-ldc-errors

Thank you for your help. Can I solve this image problem by reducing the value of s? And maintain the resolution of the output image at 1280x720. At present, I set the value of s to 2 when obtaining the mesh.txt file. Can I change it to 1?

BTW, you need to be clear on what output view is required from your system requirement.

1. There is no distortion (the most important), which can be understood as the part without the red line circled;
2. The perspective after distortion removal should be as large as possible;

Regards,

Yueqian

qian yue said:

1. There is no distortion (the most important), which can be understood as the part without the red line circled;
2. The perspective after distortion removal should be as large as possible;

Hi qian,

I suppose these are your requirements.

In real systems, view point and view size must be calibrated after cameras are mounted.
In that process, input image border (red circles) must be taken care of properly.

Okay, may I ask if after distortion correction, objects that are actually closer will appear far away from the corrected video? Can I adjust this through any parameters?

qian yue said:

if after distortion correction, objects that are actually closer will appear far away

I am not sure what you mean exactly.

It could simply be the perspective distortion of the scene.
Correction requires some assumption/estimation of the object distance like in 3D surround view systems.

There is an important note in the Introduction section of the 3D Surround View Application: it is requires to first run the vx_app_srv_calibration prior to running 3D SRV in order to have a properly calibrated image.

May I ask if running 'vx_app_srv_calibration' in advance is only to obtain the 'CALMAT. BIN' file? Can I do this through 'main. exe' in the './exe_out/.' folder instead of running 'vx_app_srv_calibration'.

Regards,

Yueqian

Hi quian,

I am not familiar with 3D surround view demo details.
Please start a e2e thread with proper title for your questions on that.

Depending on the system, 3D surround view may require GPU or depth sensors.

OK，Thank you!

Regards,

Yueqian

Hi qian,

I will close this thread if you have no further questions.

Please fell free to open a new thread if you have any more questions.

OK, thank you very much!

Regards,

Yueqian