You are on page 1of 13

FLYTBASE HACKATHON

ALPHA-Q

SUNDAR SRIPADA
&
SRIRAM VASAN
FLYTBASE HACKATHON

WAREHOUSE MANAGEMENT
ABSTRACT

To build an autonomous unmanned aerial vehicle (UAV)
and to simulate the detection process of multiple objects
in a warehouse.
Need of Warehouse Management

Today’s inventory management requires workers to scan items manually, which is a
very time consuming and an error-prone process. It’s impossible to keep track of all
items in the warehouse manually.

Thus, special aerial drones using optical sensors (cameras) which can help locate
an item in a warehouse or scan the respective RFID tags at a distance of tens of
meters are a necessity.

In larger warehouses, a drone system can be used to gauge inventory levels and
transmit the data directly to the warehouse management system. This will
facilitate the fast detection of individual items and prevent inventory mismatches.

An RFID-reading drone can pinpoint and count tagged inventory stowed inside
stand-trailers making it invaluable for large distribution centers with outdoor
goods-yards, like Walmart.
Implementation

Our project focuses on detecting multiple markers
which act as different inventory items in a
warehouse.

The detection process is done by using the drone’s
inbuilt camera to get a feed of the simulation
environment.
Implementation

This image feed is then passed through a Canny Edge
Detector to get the edges of the markers, which looks like
this:
Implementation

This edge image is morphologically transformed for the purpose of
identifying perfect contours.

This is done by performing erosion followed by dilation (called
opening, if perfomed in this order), which removes unwanted noise
from the edges image. The eroded and dilated images are shown:
Implementation

Finally, contours are detected in the eroded image. This
ensures that multiple markers are detected from the
same image and these markers are highlighted (in green)
as follows:
Demonstration

The previous slide shows multiple markers being detected and their
contours found.

In our implementation, we used Template Matching to detect the
markers.

This is done by using one marker as a template and checking if this
template is present in the camera feed image.

The two markers are successfully detected by using the
cv2.TM_CCOEFF_NORMED() method.
Advantages of Template Matching

Since only one template is tested with the image at a
time, the algorithm runs fast.

Accurate detection of template of any size can be
obtained if orientation and scaling are carefully kept in
check.
Disadvantages of Template Matching

The first disadvantage is that if we’re looking for dynamic
features, it’s better to use other techniques.

Secondly, template matching provided by OpenCV
doesn't let us check for rotations and scalings. If the P
(template present in image) was rotated by 90 degrees,
the current program would never find it.
Scalability

If a robust feature extraction algorithm such as ORB,
SURF or SIFT is implemented, the current model’s
accuracy could be manifold improved.

The resultant algorithm can be implemented in a physical
model

This can be used to ease the inventory management of
any warehouse
THANK YOU

You might also like