Head detection based on 21HT and circle existence model
Min Zhao, Dihua Sun, Yi Tang, Heng-pan He · 2012
A novel method for head detection was proposed in video sequences captured with fixed vertical mono-camera, which integrated hough transformation, hair-color distribution model and circle existence model. Target area was firstly detected using fast gradient hough transformation (FGHT). In order to overcome head area mis-detection and incapability of locating head area introduced by FGHT, hair-color classification was used to filter the candidate targets through modeling hair-color distribution. Furthermore, based on non-parameter probability theory, the probability of circle existence model was established, which finalized the stages of head detection by locating the head. Compared with average circle detection algorithm, experimental results indicate that the proposed head detection algorithm can eliminate false targets and greatly increase accuracy.