Capturing full resolution image when object is detected

vital

I am using the Oak-1 Lite. How do I change the sensor resolution to be full 13K in order to capture a full resolution still when a specific object is detected? The examples ColorCamera and rgb full resolution saver do not illustrate how to change sensor resolution on the fly. Thanks.

erik

Hi vital ,
You can't change resolution in runtime. You would neet to set resolution to 13MP, then use eg. ImageManip to
downscale isp output before streaming the video to host. Pipeline idea below. Thoughts?
Thanks, Erik

vital

Thanks for the reply. I want to do inference on device using a yolov5 model with one class. When it detects the class it would send an image at full res to the host. I would also like to have the exposure and sensitivity at a few different settings for each image grab. I am working on training a model to be as lightweight as possible in order to do this in real time.

So, I don't really need to send a downscaled image to the host -- I want to be able to do inference on device and send full res stills to the host. Is that possible with 13MP sensor resolution?

erik

Hi vital ,
Yep, that's possible. So instead of sending downscaled frames to host, you can do the inference, then use Script node to capture the still image, like in the example here: https://docs.luxonis.com/projects/api/en/latest/samples/Script/script_camera_control/#script-camera-control
Thoughts?
Thanks, Erik

vital

That's great. I am still learning aspects of this system and am have never programmed python before (the environment stuff is really getting to me…). I will try and get something working and come back with results.

Question: Is it possible to get still images captured in sync with the detections or will there be a lag between 'frame detected' and 'still captured'? Also, is there a mechanism for determining if the object has already been captured, or must we get a new still with every detection instance?

erik

Hi vital ,

You can use software syncing (I'd suggest seq num syncing): https://docs.luxonis.com/projects/api/en/latest/tutorials/message_syncing/

You can do object tracking, to only capture a single frame for each unique object (unique id): https://docs.luxonis.com/projects/api/en/latest/samples/ObjectTracker/object_tracker/#object-tracker-on-rgb

Thanks, Erik