Identifying the direction of an object from an image

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
,
1

Hello, I’m interested in determining the direction of a specific object within an image. Once I’ve identified this particular object in the image, I aim to understand its direction. My goal is to use this information to orient a servo motor accordingly. Is there a way to do this by any chance?

3

Hi,

Do you have a dataset for the “special object” with every direction beforehand?
If yes, this is similar to a template matching problem.

You can find some information with OpenCV.
For example An improved template matching with rotation and scale invariant - #4 by den8017981 - Python - OpenCV

Thanks.

1 Like
4

First of all, thank you for replying back! Maybe I will explain the object so it will let you understand more the problem I’m facing. Let’s say I have a belcony and a pigeon fly to there. What I want is that my device will identify the pigeon (that I know how to do with the image classifier with Jetson) and once it did recognize it what it need (and that is what I don’t know how to do with Jetson) is to direct my device directly towards the pigeon using servo motor. Is there any way for me to mange doing that using Jetson?

5

Hi,
Do you know if any model can do this? We have demonstration for object detection, such as ResNet or Yolo. If there’s existing model for identifying direction, it should work by running the model in DeepStream SDK.

6

The only thing is missing is exactly that- identifying direction of the object.

7

Hi,
This would need other users to share experience. It looks like there is no public model which demonstrates this use-case. Probably there is 3rdparty doing this and can share further information.

For running deep learning on Jetaon Nano, we would suggest use DeepStream SDK. You can try sample models. And replace with your own model for a try.

8

“is to direct my device directly towards the pigeon using servo motor.”

I am not sure how much of it I understood but, can’t you calculate the 3d coordinates of the detected object and then use ros (say: pose ) and use xyz and orientation to move whatever you want to move?

Sorry if I misunderstood something.

10

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