topic Re: Object tracking with Movidius and TensorFlow model??? in Intel® Distribution of OpenVINO™ Toolkit

Object tracking with Movidius and TensorFlow model???

idata — Tue, 16 Jan 2018 19:20:25 GMT

I want to track objects which i

detect with TensorFlow and Movidius stick.

Because of this i need the coordinates (f.ex. a bounding rectangle)

of the detected object.

How can i do it?

Any idea?

Thanks for help :-)

Re: Object tracking with Movidius and TensorFlow model???

idata — Wed, 17 Jan 2018 01:28:20 GMT

@hisun Please see my comment at https://ncsforum.movidius.com/discussion/564/tensorflow-models-with-object-detection#latest

Re: Object tracking with Movidius and TensorFlow model???

idata — Mon, 22 Jan 2018 22:54:50 GMT

Hello Tome,

i guess inside your SDK there must be something like a enumeration of all

objects (ID, confidence, boundings, ….etc.. ) the engine recognizes. Isn't it?

Re: Object tracking with Movidius and TensorFlow model???

idata — Tue, 23 Jan 2018 00:07:03 GMT

@hisun Usually the categories of detectable objects are contained within a categories file or the run script itself (but not limited to just this method). If you inspect the Tensorflow Mobilenets example in the ncappzoo @ https://github.com/movidius/ncappzoo/tree/master/tensorflow/mobilenets, you can see the categories of all the images the model was trained with.

For more information on bounding boxes, you can check out how TinyYolo's bounding boxes work @ http://machinethink.net/blog/object-detection-with-yolo/.