2. Outcomes of this talk
● Become familiar with
some Concepts and
Keywords
● Make you aware of some
cool applications
● Vision is a quickly
growing space, there are
ample opportunities for
our customers if only we,
and they, know about it.
3. About me and my interest in Computer Vision
2011 - Brigham Young University BS Bioinformatics
2016 - Georgia Tech MS Computer Science.
Machine Learning => computational photography => computer vision and robotics
“Have Fun Computing with Images ” Dr Essa
3 years in Vision startup building vision systems. => Aviture (July 2019)
4. Computer vision projects, I have worked on
Taters - Classification
Shot detection
Shrimp
Fish
Cookies
Foreign Material
Signature Detection
Current - putter swing analysis
5. Blending Project at GT
Here is a cool blending project using the
Gaussian/Laplacian to break apart two different
images along with a mask and then blend them
back together.
7. What is Computer Vision
Computer vision is an interdisciplinary scientific
field that deals with how computers can be
made to gain high-level understanding from
digital images or videos. From the perspective of
engineering, it seeks to automate tasks that the
human visual system can do - Wikipedia
Computer Vision tries to mimic human vision
system, of interpreting visual data.
⅔ brain used for our vision system …
8. History of Computer Vision How it evolved
In the late 1960s, computer vision began at universities
which were pioneering artificial intelligence. It was meant
to mimic the human visual system, as a stepping stone to
endowing robots with intelligent behavior. In 1966, it was
believed that this could be achieved through a summer
project, by attaching a camera to a computer and having it
"describe what it saw"
Initially some folks really thought about it as reverse Computer
Graphics…. We have found that is not really the right way to
think about it.
10. Use Cases
Computer Vision is about giving us information
from a scene.
● Classification, Detection, Segmentation,
Tracking
● Robotics, Data analytics, Shopping,
Medical, Forensics. ….
Computation photography uses that information
to make a new novel artifact.
● New Novel Artifacts
○ Panoramas
○ Colorization
○ Google Photos Stuff.
13. Stereo Vision
Hold two pencils in front of your face.
With both eyes open touch them together,
Close one eyes and reset, 2 eyes give depth perception
The latter with one eye is much more challenging.
We can mimic that by two calibrated cameras, or sub.
14. Examples of Stereo Vision Applications
Portcas - if you heard of that uses stereo vision
X-box Kinect
23. Mask R CNN
Car detection in a parking lot
Customer wanted to see if we could monitor
parking lot usage throughout the day.
First step is can we detect cars.
So with pre-trained algorithm on a large
existing COCO data set
24. Deep Learning
Imagenet challenge in 2012 considered to be the
beginning of the deep learning revolution.
They are efficient running on GPU’s 2006 is when
CUDA became a thing.
They are great cause the handle huge amounts of
data.
Very my a CNN but with lots and lots of hidden
layers.
Feature extraction and classification in the same
model.
AlexNet -> VGGNet ->GoogLeNet -> SegNet -> DenseNet Tons of Others
26. OpenCV (Open Source Computer
Vision Library) is an open source
computer vision and machine
learning software library
C++, python or Java bindings
Great tutorials. Lots of really cool
things you can do
27. Other Tools
Dlib C++ library
Scikit Learn
Matlab
numpy/scipy
Many Others….
Computer Vision: Algorithms and Applications, 1st
ed. http://szeliski.org/Book/ :
https://www.learnopencv.com/
https://www.pyimagesearch.com/2018/07/19/opencv-tutorial-
a-guide-to-learn-opencv/
OpenCV.org Documentation is pretty awesome.
28. Deep Learning
Keras, Tensorflow, Caffe and others
Publicly available image datasets: (google them)
like COCO (common object in context)
Being used to solve a lot of interesting problems
Cloud Platforms on AWS, Microsoft, Google etc.
Free access to some CUDA Cores
https://colab.research.google.com/