Observing Malaysian Social Media

Himpunan Kebangkitan Rakyat #KL112 Crowd Estimate

Himpunan Kebangkitan Rakyat (People’s Uprising Rally / #KL112) was a protest rally held at Stadium Merdeka, Kuala Lumpur on January 12th, 2012. It was organised by Pakatan Rakyat and various NGOs.

At its peak the crowd is estimated to have been between 63,976 – 78,193 people. This number is not expected to be revised, unless some new evidence surfaces to suggest a different crowd size.


Photos and videos that showed the size of the crowd were evaluated. These were used to determine the extents of the area covered by the crowd and the crowd density. The area was then refined by removing obstacles and estimating the percentage of space used. A crowd density of 4 – 12 sq.ft. per person was used.

Building the polygons

After initial evaluation of the photos, one was chosen as a reference to build the polygons from. The polygons were then adjusted after evaluating photos and videos. The steps are shown in the diagram below:


  1. The original photo
  2. Satellite photo, perspective-adjusted
  3. The original photo, perspective-adjusted to match the satellite photo
  4. The completed polygons on the map


The polygons in the map below shows the extents of the crowd, divided into different zones. The total space covered by the polygons is 456,627.05 sq. ft. The space taken up by the crowd is 275,532.03 sq. ft. These measurements do not include the stadium seating.


The estimated crowd in each zone is:

A = 20,122 people

B = 1,188 people

C = 8,652 people

D = 2,474 people

E = 7,428 people

F = 1,221 people

Stadium seating = 30,000 people

Total = 71,085 people

Margin of error is +/- 10%.

The estimated size of the crowd is 63,976 – 78,193 people.

Additional reference photos

These satellite and perspective photos of Stadium Merdeka were used to measure the fixed area of the stadium.


Some photos of the crowd:

20131121134369373420 rally2q 185742_386061841488913_935269686_n

Some shots from a distance:

603049_414373715309952_439321023_n  227642_10151235202583845_316404824_n


  1. The shaded areas on the Google Map are there to give an idea of the area used in calculation. It is not drawn with the actual latitude/longitude coordinates used in our maths, which was a more detailed polygon.
  2. Calculated area total does not include buildings. A more detailed polygon was used for calculation. This polygon could not be exported for use in Google Maps, so an approximate shape was redrawn.
  3. Buildings are at an angle in satellite images, so lines that appear to be going through are actually going through an open area next to it.
  4. All figures in this post are estimates, and not expected to be 100% accurate.

3 Responses

Subscribe to comments with RSS.

  1. Thanks for doing this and sharing it.

    The technique is fairly self-evident, and therefore attractive. I’m curious to know how you arrived at the error percent.
    I’m wondering whether there are some more objective measures, e.g.

    (1) Comparing counts of unique ticket scans at football matches with results derived from the application of crowd estimation techniques such as the one above, or

    (2) lab simulations, e.g. differences in estimates of the same number of beans spread multiple times on “stadium-like” shapes or

    (3) results from different analysts applying the same techniques to the same data sets.

    Thanks again.

    rama ramanathan

    January 16, 2013 at 12:51 pm

    • Hi Rama,

      Thank you for your comment. Ticket scans would be better. Lab simulations sounds difficult and expensive – you would need very small beans or a big stadium mockup. From what I’ve read before about crowd estimation in the US, organisers employ multiple analysts and take an average from that. One effective method using surveillance balloons is described here: http://www.cbsnews.com/8301-503544_162-20015214-503544.html . Basically they take shots from multiple angles and overlay a grid on it to calculate crowd density.

      For this rally we’re working with less detailed images so its tricky. You can only fit so many people within a specific area, and free access to satellite images makes this a viable option. Compared to the traditional grid-based technique, where your estimate may be influenced by optical illusions created by the angle of the photo. You can mitigate that with more photos from more angles but combining both techniques should yield the best result.




      January 16, 2013 at 2:59 pm

      • Hi Kamal,

        Thanks for responding.

        I’m looking forward to reading more of your posts.

        It’s good to see a result presented together with the approach used to obtain it, and the corresponding estimates of error. I hope there are enough journalists in Malaysia who can use work like yours to ask appropriate questions when they are given crowd numbers by the Malaysian officials or politicians.


        rama ramanathan

        January 17, 2013 at 12:40 am

Comments are closed.

%d bloggers like this: