I have been trying to setup some triggers based on the count of people in a room, however the output from tensorflow tends to be very noisy… It frequently gets hits and then doesn’t or jumps to one higher or lower in count.
I have tried various filters but it seems I need to average out the output so much that the time delay would be too high.
Anyone have tips? I have tried adjusting the sensitivity so that it produces fewer zeros but If I take the cut off below 70 % certainty the number of objects detected as people jumps substantially.