Hi,
I am trying to use fastcv fcvCreateBGCodeBookModel and related APIs for foreground detection. The qulaity of foreground detected in my implementation using the FastCV implementation is bad compared to OPenCV 2.0 implementation. The openCV 2.0 implementation is same as https://code.ros.org/trac/opencv/browser/trunk/opencv/samples/c/bgfg_cod....
The FastCV implementation for codebook does not provide a way to set the color thresholds as below
bg_model->cbBounds[0] = bg_model->cbBounds[1] = bg_model->cbBounds[2] = 10;
*/
When I run the code snippet at the link below , after replacing OpenCV 2.0 calls with FastCV calls, the quality of foreground detection goes bad after the first 10 frames(post training). 15 frames are used for training, while stale threshold was 7.
https://code.ros.org/trac/opencv/browser/trunk/opencv/samples/c/bgfg_cod...
I have tried this implementation on Nexus4(Snapdragon s4), Xperiapro\(Snapdragon s2) and NexusPrime devices
Thanks in Advance
Hi,
Is the performance issue caused by the different bg_model configuration in OpenCV and FastCV?
Thanks,
Lei
Lei,
I am not aware of a possible way to configure bg_model in FastCV. If you are aware of one, please let me know.
API in FastCV
API in OPENCV 2.0
CvBGCodeBookModel *cbmodel= cvCreateBGCodeBookModel();
Thanks,
Arun
Hi Arun,
I'm so sorry that currently there is no API to customize the configuration.
Do you really need to change to model setting for your case, otherwise it will lead to the foreground detection issue that you mentioned?
Thanks,
Lei