You are on page 1of 3

Minor Bug Fixes and Improvements

PATCH NOTES
As we have seen in the timeline established, we have allocated the
final review for bug fixes and implementation of certain improvements
that we are suggesting by the professor in charge, Dr G Sakthivel
An allusion to the recognition pipeline, which implied that the range
of detection must be increased to facilitate a robust system was taken
into consideration and implemented thusly
So how did we go about doing it?

We thought about training the images in grayscale instead of


training them in RGB instead of writing a code to convert it every time,
but when we loaded the image in opencv, it so happened that the domain
spaces and the edge parameters turned to zero as the array sizes were
different in RGB and grayscale
This was the problem, and that is the reason the detection range
was contracted into a very small range with highly efficient detection
happening only when the image was perpendicular.
After training it with RGB and obtaining the new weights, the range
was expanded
Also as a side note, we tried to train different angles separately , but
it took a long time as aan abundance of angles, and if the angles were not
precise it didn’t detect the license plate at all.
After the weights were recalibrated to the new filter, we can then
proceed to the recognition pipeline
To understand it a bit better, look at these images.
Minor Bug Fixes and Improvements

Before recalibration
Minor Bug Fixes and Improvements

After recalibration

You might also like