Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

OBJECT RECOGNITION FOR NAVIGATION

FINAL PRESENTATION
PLANETARY EXPLORATION AND SPACE ROBOTICS I

DAVID BENAVIDES

MSE WINTER SEMESTER 2019-2020


Agenda
⚫ Approach & Results
⚫ Next steps
Approach & Results
⚫ Finding coordinates of the detected Time: 26s → Too high!
Compared with tiny YOLO = 1 second,
object → Slight YOLOv3 code [1] but no so accurate.
modification.
Obtained Information after running the code:

2D coordinate
system used in
the image
Detected objects after running YOLOv3 full weights. processing.

Objects coordinates [2]


Approach & Results
⚫ Storing coordinates, name, size and middle points in a txt file → YOLOv3 code
modification. Obtained Information after running the code:

Generated ImageCoord.txt file:

Detected objects after running YOLOv3 full weights.


Approach & Results
⚫ Real time detection and distance estimation (using tiny Yolo) with single camera →
YOLOv3 code modification.
• Triangle Similarity
Method.

• Estimated known
size of the object to
detect.

• Calculate the
“Focal Length” of
the camera.

• Knowing the above


parameters, the
distance can be
estimated.
Next steps
⚫ Link YOLOv3 (modified) with ROS.
⚫ Incorporate Intel Real Sense.

[3]

⚫ Keep working with distance estimation from camera to


object.
⚫ Train YOLOv3 according to ERC needs.
⚫ Use a GPU to improve running speed of the algorithm
Thank you!

Questions?
References
[1] https://pjreddie.com/darknet/yolo/

[2] https://www.pyimagesearch.com/

[3] https://www.pyimagesearch.com/practical-python-opencv/?src=sidebar-single-
weekend

[4] https://www.pyimagesearch.com/2015/01/19/find-distance-camera-objectmarker-
using-python-opencv/
Image references
[1] https://pjreddie.com/darknet/yolo/

[2] https://answers.opencv.org/question/31470/the-coordinate-system-of-pinhole-camera-
model/

[3] https://www.conrad.com/p/intel-realsense-depth-camera-d435-full-hd-webcam-1920-x-
1080-p-clip-mount-stand-1707630

You might also like