Skip to content

Latest commit

 

History

History
89 lines (63 loc) · 2.74 KB

README.md

File metadata and controls

89 lines (63 loc) · 2.74 KB
Raw

UberPoC

Description

UberPoC is an autonomous software able to detect lines on the ground, detect signs, traffic lights and pedestrians.

Installation

 $ git clone https://github.com/PoCFrance/UberPoC
 $ cd UberPoc

Python package dependencies:

  • pyglet
  • imutils
  • numpy
  • matplotlib
  • opencv-contrib-python
  • tensorflow
  • scikit-learn
  • keras
  • beautifulsoup4

Quick Start

If you want to try Line Detection System through a video or the Duckietown Simulator :

$ ./run.py --video-name [video.mp4]
or
$ ./run.py --duckietown

You can press L to toggle line detection
You can also use N to see different step of normalization

If you want to try Sign Detection over an image or your Camera:

$ ./run.py sign_detection --cam
or
$ ./run.py sign_detection --img [img.png]

To try Human Detection over an image:

$ ./run.py human_detection --img [img.png]

Features

Implemented

  • Line Detection : Displays the lines on the ground while the car is driving.
  • Pedestrian Detection: Circle the pedestrians and cyclists on an image.
  • Detection and recognition of road signs : Circle and name traffic signs

Future

  • Auto pilot : Combine all the features in order to have an autonomous car
  • More road signs and traffic lights : Find dataset of others signs and traffic lights to train AI

Description of each features

Line Detection

  • We cut the top of the image in order to gain precision by having only the bottom with the lines
  • Then we apply a red filter on the image to better differentiate the white lines from the rest of the image
  • Using the Canny function of OpenCV, the lines are cut
  • Using the HoughLinesP function we get an array with the different points that make up the lines
  • With these arrays the lines are estimated and displayed on the screen

Pedestrians Detection

  • To detect the pedestrians on a road we use a function of OpenCV (HOGDescriptor_getDefaultPeopleDetector()) who get all the regions of each person on an image
  • Then we iterate on these regions and display a rectangle around each pedestrian

Traffic Signs Detection

  • To detect the different traffic signs present on an image or video, we use an AI based on a particular model the "Faster R-CNN" (Region-based Convolutional Neural Networks)
  • AI will scan the full image of small regions to be able to detect a sign on the image
  • Once the sign is found it will identify it and then surround it on the image with its accuracy rate and type

Authors