You are on page 1of 13

WP: OBJECT RECOGNITION

FOR NAVIGATION
MID-TERM PRESENTATION
PLANETARY EXPLORATION AND ROBOTICS I
DAVID BENAVIDES
MSE WS 2019

17/12/2019
AGENDA

• Theoretical background (Investigation so far)


• First hands-on trials & Results
• Next steps

D. Benavides Object recognition for navigation WS 2019


THEORETICAL BACKGROUND
• Object recognition Object detection
An object recognition An object detection
algorithm identifies which algorithm not only tells you
objects are present in an which objects are present in
image. It takes the entire the image, it also outputs
image as an input and bounding boxes (x, y, width,
outputs class labels and height) to indicate the
class probabilities of location of the objects inside
objects present in that the image. [3]
image [1]. [2]

Image classification Object Localization

Predict the type or class of an Locate the presence of objects in an image and indicate their
object in an image. location with a bounding box.
Input: An image with a single
object, such as a photograph.
[4] Input: An image with one or more objects, such as a photograph.
Output: A class label Output: One or more bounding boxes (e.g. defined by a point,
[5]
width, and height).
D. Benavides Object recognition for navigation WS 2019
THEORETICAL BACKGROUND
• Books reviewed

[6] [7]

D. Benavides Object recognition for navigation WS 2019


HANDS-ON TRIALS & RESULTS

[9]
[8]

Train your first neural network

on a custom image data-set Introduction to the problem of
object recognition and state-of-
the-art deep learning models
designed to address it.

D. Benavides Object recognition for navigation WS 2019


HANDS-ON TRIALS

R-CNN YOLO (You Only Look Once)


R-CNN

YOLO

Fast R-CNN

YOLO v2

Faster R-CNN

YOLO v3

Introduced in 2013 by Ross Girshick. 
To help increase the speed of deep

R-CNNs are one of the first deep learning-based object detectors, YOLO
learning-based object detectors and are use a one-stage detector strategy
an example of a two-stage detector.

First introduced in 2015 by Joseph

Problem: Speed. R-CNN was incredibly Redmon.
slow, obtaining only 5 FPS on a GPU. [10] 
Real-Time Object Detection, obtaining up
to 30 FPS on a GPU. [11]

D. Benavides Object recognition for navigation WS 2019


HANDS-ON TRIALS
• Tutorials YOLO
2nd Example

[12]

1st Example

D. Benavides Object recognition for navigation WS 2019


HANDS-ON TRIALS
• Tutorials YOLO
3rd Example

D. Benavides Object recognition for navigation WS 2019


HANDS-ON TRIALS
• Tutorials YOLO
4th Example

D. Benavides Object recognition for navigation WS 2019


NEXT STEPS
• Trials YOLO for real time detection (video).
• Try to adapt YOLO to ERC challenges (own data training, code modification):

Rocks Detection

Landmarks Detection

Persons Detection

D. Benavides Object recognition for navigation WS 2019


Thank you!

Questions?
REFERENCES
1- https://www.mathworks.com/solutions/image-video-processing/object-recognition.html

3- https://www.mathworks.com/solutions/image-video-processing/object-recognition.html

4- HTTPS://MACHINELEARNINGMASTERY.COM/OBJECT-RECOGNITION-WITH-DEEP-LEARNING/

5- https://machinelearningmastery.com/object-recognition-with-deep-learning/

10- HTTPS://WWW.PYIMAGESEARCH.COM/2018/11/12/YOLO-OBJECT-DETECTION-WITH-OPENCV/

11- HTTPS://WWW.PYIMAGESEARCH.COM/2018/11/12/YOLO-OBJECT-DETECTION-WITH-OPENCV/

D. Benavides Object recognition for navigation WS 2019


IMAGE REFERENCES
2- https://www.mathworks.com/solutions/image-video-processing/object-recognition.html

6- https://www.oreilly.de

7- HTTPS://WWW.OREILLY.DE

8- https://www.pyimagesearch.com/2018/09/10/keras-tutorial-how-to-get-started-with-keras-deep-learning-and-python/

9- https://machinelearningmastery.com/object-recognition-with-deep-learning/

12- HTTPS://PJREDDIE.COM/DARKNET/YOLO/

D. Benavides Object recognition for navigation WS 2019

You might also like