You are on page 1of 2

ENGO 559/659 Count Tents in a Refugee Camp Project 2

Lab Sessions: February 6, February 13, and February 27, 2024


Report Due (no penalty): Fri, March 1, 2024 at 4:30 pm
Final Deadline (with penalty): Mon, March 4, 2024 at 11:59 pm
Marks: 25% of course grade

The United Nations High Commissioner for Refugees (UNHCR) estimates that there
are 65.3 million forcibly displaced people worldwide. Of these 21.3 are classified as refugees,
i.e. they are forced out of their own country. In many cases large numbers are forced to
flee their country across the border to a neighbouring country. Refugee camps are set up,
and it is always a struggle to house, feed and service large numbers of people in a short
period of time. In the early stages of a refugee camp, the authorities do not know how
many tents and how many people are in the camp.
The objective of this project is to detect and count tent dwellings in a refugee camp.
You may choose one of two camps: the Lukole camp in Tanzania, set up for refugees fleeing
the genocide in Rwanda in April 1994, and Camp Mille in Chad, set up for refugees from
the war in Darfur, Sudan, which started in 2003.
The data for each camp consists of high resolution remote sensing image data collected
by the IKONOS and QuickBird satellite sensors. For each camp, you are provided both
the panchromatic channel, and four multi-spectral channels. You should do some research
to find the characteristics of the data, e.g. centre and width of the spectral channels, and
the spatial resolution (which is different for the pan versus the multi-spectral channels).

Lukole camp in Tanzania: Date: September 24, 2000; Sensor: IKONOS

Camp Mille Chad: Date: June 12, 2004; Sensor: QuickBird

The objectives are: first write an algorithm to detect the tents as objects; then count
the detected objects. You may use either the panchromatic channel or the multi-spectral
channels for detection. In order to assess the performance of your detection/counting
algorithm, you must have something with which to compare your algorithm-count. You
will perform a manual count of the tents in the image you have chosen. Not having been
on the ground in either camp, this might be a challenge in some cases. Do your best to
count all image objects you believe are tents. Since there are too many tents to reasonably
count in the full image, you should crop the image to a smaller size. Note that the spatial
resolution of these two datasets is different, but the tents tend to be more spread out in
Camp Mille. Thus, I would like you to crop the images so that there are no fewer than 100
tents in the cropped image.
Two common performance metrics for object detection are: errors of omission (missed
targets) and errors of commission (false targets). You should calculate both of these. The
tents have different visual appearances with the images. For example, most tents are bright
in the Lukole image, but some have a darker appearance. Make an effort to detect all tents.

1
ENGO 559/659 Count Tents in a Refugee Camp Project 2

You will have to design different detection strategies for the different tents. Analyse the
detection performance on all tent types.

You might also like