I tried running the C++ detection and tracking sample as described here while using the live camera instead of a prerecorded file, but the camera doesn't seem to detect or classify any moving vehicles or pedestrians. I will only see bounding boxes get drawn when I lower the confidence threshold to 0.1 by setting --detector-confidence-threshold 0.1 at runtime. However, this of course causes many bounding boxes to get drawn on the screen and incorrectly classify objects that aren't vehicles or pedestrians. Is there something I need to do to get this model to work while using the live camera feed instead of a prerecorded file?