RTOS/TDA2HV: Stereo cam problem

Hello Ravi,

It seems left and right cameras are not well rectified. Please make sure that

1) you run stereo calibration use case first. Also low-distortion cameras are recommended (I see barrel distortion for left camera input). Otherwise, calibration may not come up with correct rectification tables.

2) you connect right camera to the 1st camera port (FPD link cable) and left camera to the 2nd camera port.

3) 10 ~ 15 cm of baseline is recommended

I also put some notes in the attached documents. Please check and let me know if you have any questions.

TDA2x_Stereo_Usecase_Notes.docxTDAx_Disparity_Calc_Dim.xlsx 

Hi Sir,

i understood most of things. but

2) you connect right camera to the 1st camera port (FPD link cable) and left camera to the 2nd camera port.
is it necessary ?? can you please elaborate this??

Thanks,

Ravi.

For example, if you use UB914 deserializer, you can connect up to 6 cameras to it as shown in the below figure. You have to connect the right camera to Cam 1, and left camera to Cam 2.

hi Sir,

We are using separate deserializer for each camera. in that case what i have to do??

now i have set baseline to 10cm.

still not correct output.

Thanks.

Hi Sir,

1. i ran calibration usecase as you suggested in document.

At first use case was taking table from .c files now it is taking from QSPI flash.

2.  i am attaching screen shot of stereo calibration use case

i am getting too big output of right camera (very zoom)

Hi Sir,

by changing camera we got correct output. but still there is very much noise in output. i did calibration many times but i am not unable to remove noise from output. otherwise it is working with objects.

Thanks,

Ravi.

Hi Sir,

I have run auto calibration use case and got output is like:

we got red dots on screen only when i select 2. capture stereo from menu:

is there any possibility of errors in use case/algorithms? is it fully tested?

For this natural image based calibration, you try to align left and right cameras well enough. Otherwise, it cannot find corresponding points (shown in red dots - you should see red dots when you select 1 and 2 from the menu.

First of all, your stereo cameras are not well aligned. For example, the right camera looks rotated (roll angle), and zoomed in. If necessary, you can try with shorter baseline between left and right. And the lens distortion may cause calibration failure.

When you collect feature points (by selecting 1 and 2 from the menu), put something in front of camera so that it can capture distinct features. But do not use checkerboard chart (or something has repeated pattern), because it can fool feature detector.

When you select 3 from the menu, it create calibration table. You can see from a console the rectification error value. This error should be around 0.2 (shouldn't be larger than 0.3) 

If you continue to have the same issue, I encourage you to try offline calibration.