search

Idein / actcast-support

13 stars 0 forks source link

Local pose estimation #7

Open feymartynov opened 4 years ago

feymartynov commented 4 years ago

I tried Actcast's pose detection demo on my RPi Zero using the tutorial and it worked: I got a photo with detected pose on it. However I'm confused on what to do next. What I would like to do for my project is to analyze the pose locally with at least 2 or 3 FPS (I saw you got even 6 on Zero in your demo), detect specific gesture that the person makes and trigger a motor through RPi's GPIO. I have a Python script that does all the stuff. All I need is to get keypoints from a Videocore accelerated pose estimator. The device works offline by the way. Though Actcast can do a lot more things like orchestrating multiple devices, sending HTTP hooks and so on it's not clear to me if it could help me with this relatively simple task? Is there a way to get keypoints from the pose estimator act and process them locally?

notogawa commented 4 years ago

Thank you for trying Actcast.

The Pose Estimation app is just a performance demo now. This app doesn't send detected keypoints to outside RPi. We consider a plan to impl additional features to stream keypoints using WebSocket.

If you want to develop your original app using our VideoCore4/6 acceleration technology, you should join our partner program. Now, we provide Actcast SDK to our partner.

feymartynov commented 4 years ago

Is there a way to use these keypoints in a custom script inside RPi?

notogawa commented 4 years ago

To be our partner and use SDK is the only way to develop a custom application for now.

surfertas commented 4 years ago

Is there a way to use these keypoints in a custom script inside RPi?

Not sure if its the best method but I used pythons websockets library to send serialized data over the network for consumption on client-side.