Training a CNN and using OpenCV

Hi:

I have to do a project to detect yellow lines on the road using a CNN.I dont know which direction take to do this. I have think i can do the following:

-First train a CNN using DIGITS or any other tool. (I dont know if i should create a new CNN or retrain other CNN like GoogleNet). I thinking on using the Caffe framework.

-Then download to the Jetson TX2 the .prototxt and .caffemodel files of the trained CNN.

-Make a C++ program or Python using OpenCV and use the files that I download before. Use the on-board camera for video input and give the frames to the input of the trained CNN.

It’s possible to do this? If you have a better way can you tell me what is it?

Thanks for help!

Hi,

You can check our tutorial to get some idea:
https://github.com/dusty-nv/jetson-inference

As you said, you will need to train a model for your use-case and inference it with the above sample.

By the way, we have LaneNet to detect road line in DriveWorks SDK.
https://developer.nvidia.com/driveworks

Although it doesn’t support Jetson, you can try it on your host to have some direction.
Thanks.

Hi, thank you for the info.

A few days ago I applicated for join de the drive developer program but my application was rejected because I use my personal mail for login.

It will help me if you can accept my application. I work for a company and I´m developing an Artifical Vision project. I try to edit my e-mail in the application for including my corporate e-mail but I cant.

I dont know if you can do something about that.

Thnak you for your help!

Hi,

You can use DEVZONE account to login.
The difference is the access for DRIVE AGX platform which should be okay with you.

Thanks.

hello

what is the Devzone account ??

this happened to me too , my application was rejected too .

can i have some help please

Hi,

Sorry for the late reply.
Devzone account is the account you use to login this forum.

Thanks.