Object tracking of object defined by a bounding box

DannyCheng

Hi,

We are trying to track an object, but instead of relying on one of the NN-based object detectors on the camera, we would like to pass in the object (define a bounding box?) detected on the PC side. Is this possible? I am still trying to go through the object detector documentation but any pointers would be much appreciated.

erik

Hi DannyCheng ,
That would be possible as well🙂 ObjectTracker node accepts ImgDetections message, which you can create on the PC (together with 2x ImgFrame) and send all of these to the Tracker (via XLinkIn nodes), which will track objects. Thoughts on this approach?
Thanks, Erik

DannyCheng

Yep that's the conclusion I got from the documentation, haven't figured out how to compose the ImgDetections message yet, so I'll do more reading/experimentation today and report back

erik

Hi @DannyCheng ,
We can also prepare a simple example on how to construct it, if you'll have trouble with that.
Thanks, Erik

DannyCheng

Hi erik, that would be very much appreciated if you could do that, thanks! 🙂

erik

Hi DannyCheng ,
Quick snippet:

    from time import monotonic
    import depthai as dai

    det1 = dai.ImgDetection()
    det1.label = 1
    det1.xmin = 0.1
    det1.xmax = 0.3
    det1.ymin = 0.2
    det1.ymax = 0.4

    dets = dai.ImgDetections()
    dets.setSequenceNum(1) # 2, 3..
    dets.setTimestamp(monotonic())
    dets.detections = [det1]

You can then send the dets ImgDetections message to the OAK device via XLinkIn. I hope this helps!
Thanks, Erik

DannyCheng

Thanks erik , will give this a try. I'll let you know if I run into issues with XLinkIn 🙂

Update: okay I think I'm quite lost with this XLinkIn thing, would you be able to give me some pointers on this too? Apologies for the trouble...

erik

Hi DannyCheng , Have you already checked XLinkIN docs? https://docs.luxonis.com/projects/api/en/latest/components/nodes/xlink_in/

DannyCheng

Hi erik , yes I have. I am basing my code on the object_tracker.py code for now. I am capturing ImgFrame without tracking, and pausing it and manually selecting the ROI points using OpenCV. After that, I create the ImgDetection object using your snippet, however it is not clear the parameter/variable name I need to send the ImgDetection to via XlinkIn, I cannot seem to find this in the documentation?

erik

Hi DannyCheng , what parameter/variable name are you referring to?