what are the criteria for selecting CCTV camera / sensor that shall be used for counting people?
I think this really depends heavily on the analytics application you are using, and what the recommendations are for coverage area.
Different analytics vendors will have criteria for getting optimal counts. This can include things like amount of time needed to "detect" the person. Some vendors may want wider shots to get more frames of the person moving, while others may prefer tight shots to minimize the number of people that can be in the frame at once, so there is not as much processing load.
You will likely find that for the best results you will need a camera dedicated to the counting function, mounted to get an overhead view of the entry area.
Some solutions also recommend specialized cameras, like 3VR's Time of flight camera, and I have seen others that prefer/use thermal cameras for counting applications.
What is the goal for the crowd management requirement? To count the number of people inside a building/space, detect a large crowd/group in an area, something else?