The requirement of the algorithm is to detect human shapes. While the standard HoG or Harr cascade can be used, I am exploring whether the human shape - which primarily includes face+neck+upper-torso shape as some kind of template and identify human shape.
It will facilitate human detection in extremely low light condition where practically all i one can see is a "shadow like" shape of the human.
PS: I do not have computational power (GPU) to run deep learning. So looking for traditional vision based algorithm.