• State
  • Locked Locked
  • Replies 3 replies
  • Subscribers 98 subscribers
  • Views 511 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

OpenVX object detection on TDA2XX

Luka Karan
Prodigy 190 points


I am trying to make application to recognize traffic signs using OpenVX. Can someone recomend me what  functions should I use to recognize object (for example square,triangle, circle) or even help me with some object detection examples so i can modify it?

I am using VIsion SDK 03_03_00_00 on TDA2XX chip and I started from your example for Harris corners detection.

  • 0
    TI__Genius 17580 points

    Hi,

    Did you check our TIDL based object detection use cases @ /vision_sdk/apps/src/rtos/usecases/tidl/ in the latest Vision SDK https://www.ti.com/tool/PROCESSOR-SDK-TDAX

    Thanks,

    Praveen

  • 0 in reply to Praveen Eppa1
    Prodigy 190 points

    I did, but as I can see in latest Vision SDK is used network (i do not have network access on my TDA2XX) and there are none of OpenVX node functions (for example vxColorConvertNode, vxExtractNode...) 

    I need to make OpenVX program that recognize traffic signs so I would like first to recognize some shapes and than to adapt it to recognize signs.

  • 0 in reply to Luka Karan
    TI__Genius 10365 points

    There are multiple ways to recognize shapes and there are enough literature available online. Two main approaches are,

    1. Computer vision based

    2. Deep learning based

    For either one of them you will have use some of the existing kernels in OpenVx or write your custom kernel. 

    Steps to try are,

    Narrow down on the approach how you want to recognize a shape? Deep learning approach provides better accuracy these days. Train a network to identify the shapes using TensorFlow/PyTorch/Caffe. Use the TIDL library to import a trained model. Try it standalone on EVM. Then switch to an OpenVx based application which uses TIDL.

    Regards,
    Shyam