Issues with tracking

[tomcreutz]

Hallo,

as part of my bachelor thesis I am trying to get the Tracker running right now. But I am having troubles to get it running.

Changing the scale of the bounding box is working perfectly for me. But for some reasons it does not track or tracking is working only sometimes and not really accurate.

Am I right that the image concatenate function of the ConcatenateFeatureChannelProcessor takes the vector of RGB and Depth Image Map and only returns a Vector with the Depth Image Map? Because of that the KCF-Tracker which should probably be used for actual tracking is never called.

So could that be the reason why the tracking isn't really working for me?

Thanks in advance!

[copperwiring]

@tomcreutz Are you working with a depth sensor or with a dataset? Because the code doesn't seem to have any linking with OpenNI, Kinect SDK or OpenKinect libraries for the code to work with, in real-time.

[tomcreutz]

@copperwiring I'm using it with ROS and a Kinect right now. I get the rgb and depth data via ROS topic, put them in an array std::array<cv::Mat,2> and simply call reinit for initalisation and update for updating. Thats how it's done in their tracker_run files.

[copperwiring]

@tomcreutz Oh. Okay. I am trying to use Kinect with OpenCV but since there is no a.) Kinect SDK b.) OpenKinect 3.) OpenNI library used, OpenCV can't treat Kinect as a simple webcam. Hence, I am not able to move forward. Honestly, I had thought, as real-time, it'll directly fetch Kinect video on which we can draw bounding box. But this isn't possible without a linker.

Any idea how can it be done for OpenCV and Kinect only for this code (if possible)?

P.S. It'll be helpful if you can attach a snapshot of the line via which you are sending the data.