10

enter image description hereWe need to count the number of people in a crowd and we have used the opencv_traincascade but it detects many things.

What could be the other approach we can use?

Rowland Shaw
  • 37,700
  • 14
  • 97
  • 166
user2940240
  • 159
  • 1
  • 2
  • 10
  • Could you please provide more details on whats happenning with the current algorithm? What do you mean with "it detects many things", also if you could please post a sample image of the crowd looks it would be better, the crowd-counting algorithm is very different given a different camera perspective. – adelriosantiago Nov 03 '13 at 08:17
  • ...here is the image http://i.stack.imgur.com/YMqNQ.png – user2940240 Nov 06 '13 at 10:49
  • We have different static angles of cameras. I trained using opencv_traincascade to have an xml for the head detection but the result xml detects head more than ones and even other objects present on the image. – user2940240 Nov 06 '13 at 10:57

2 Answers2

5

I was looking for something like that. I have found this project HeadCounter ( https://github.com/yquemener/HeadCounter ) and some papers: "Counting in Extremely Dense Crowd Images" ( http://crcv.ucf.edu/projects/crowdCounting/index.php ) and "Crowd Size Estimation" ( http://course1.winona.edu/cmalone/promotion/UndergraduateResearch/Cariveau%20Report.pdf )

Thiago Mata
  • 2,825
  • 33
  • 32
0

There are several ways of doing this. I have done a similar project years ago with the help of Aforge.net.

I'll suggest you to use some existing image library like emgu-cv or aforge for this task.

I think http://www.codeproject.com/Articles/239849/Multiple-face-detection-and-recognition-in-real-ti article may help you in achieving this functionality with sample project.

milan m
  • 2,164
  • 3
  • 26
  • 40
  • ...Thanks. I'll try your suggestion. The people in our image is not that clear and the camera is in static position. – user2940240 Oct 31 '13 at 15:25