Improve the efficiency of HOG for people detection
Hie,
I'm looking for ways to improve the performance and efficiency of HOG for people detection. The current peopledetector function results in quite a few false positives and some false negatives when I give an input video. Is there any method to improve it ? Or should I go with training?
The best way to improve efficiency of people detection is to go find out what is being done that is better than HOG, rather than trying to magically improve a 15-year old method.
Start with https://arxiv.org/pdf/1602.01237.pdf
Thanks for your suggestion.