Fish-eye dewarp metrics

• Hardware Platform (Jetson / GPU)
GPU
• DeepStream Version
6.1.1
• TensorRT Version
8.4.1-1+cuda11.6
• CUDA Version
11.4
• NVIDIA GPU Driver Version (valid for GPU only)
470.199.02
• Issue Type( questions, new requirements, bugs)
Question:
I’d like to know if there’s any metrics to measure the accuracy (or likeliness) of the fish-eye dewarp streams compared to normal camera output. As our application use this output stream + AI model to detect actions, the result via fish-eye output are really bad compare to normal camera with same angle…

I also want to know if there’s any recommendations about configs related to camera setup (angle. height, …) so we can apply to our case.

We re-use the dewarp example with this config:

[property]
#dewarp-dump-frames=10
output-width=1280
output-height=720
# aisle-calibration-file property is getting used in case of deepstream-360d-app,
# if set below properties under group [surfaceX] will be ignored
#aisle-calibration-file=csv_files/nvaisle_2M.csv
# spot-calibration-file property is getting used in case of deepstream-360d-app,
# if set below properties under group [surfaceX] will be ignored
#spot-calibration-file=csv_files/nvspot_2M.csv

#########################################
# Note - Max 4 surfaces are supported
#########################################R
num-batch-buffers=1

[surface0]
# 1=PushBroom, 2=VertRadCyl 3= Perspective_Perspective FISH_PERSPECTIVE=4, FISH_FISH=5, FISH_CYL=6, FISH_EQUIRECT=7,
# FISH_PANINI=8, PERSPECTIVE_EQUIRECT=9, PERSPECTIVE_PANINI=10, EQUIRECT_CYLINDER=11, EQUIRECT_EQUIRECT=12 EQUIRECT_FISHEYE=13,
# EQUIRECT_PANINI=14, EQUIRECT_PERSPECTIVE=15, EQUIRECT_PUSHBROOM=16, EQUIRECT_STEREOGRAPHIC=17, EQUIRECT_VERTCYLINDER=18
projection-type=2
surface-index=0
#dewarped surface parameters
width=1280
height=1280
top-angle=0
bottom-angle=-120
pitch=90
yaw=0
roll=0
focal-length=350
# Z axes corresponds to roll, X corresponds to pitch and Y corresponds to yaw
# Six combinations are possible : XYZ, XZY, YXZ, YZX, ZXY, ZYX
# Default is YXZ i.e. yaw, pitch, roll
# In this example camera first rolls by -38 degrees and then a pitch of 90 degree gets applied
rot-axes=ZXY

Also. we’d like to know which factor impact on the choice of projection-type.

Thank you!

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

Theoretically whe you capture the image with fish-eye camera, the content is already different to the image you captured by normal camera. The dewarpper algorithm can only help to remove the distortion as much as possible, the reconstructed content can not be exactly the same as the normal image.

There are some camera related parameters should be configured according to your camera. Dewarp Deepstream 6 - Intelligent Video Analytics / DeepStream SDK - NVIDIA Developer Forums

They are just different projection algorithms. It depends on your scenario.
panini.pdf (tksharpless.net)
Perspective_projection - Wikipedia

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.