Final Presentation - Shane Tuohy

assoverwroughtAI and Robotics

Nov 6, 2013 (3 years and 7 months ago)

101 views

Shane Tuohy


In 2008, rear end collisions accounted for
almost 25% of all injuries sustained in road
traffic accidents on Irish roads
[
RSA Road Collision
Factbook

2008]




Effective distance determination can go a
long way to reducing injuries


Mercedes Pre
-
Safe


Audi Pre
-
Sense Plus


Toyota Pre
-
Collision System



All are RADAR systems


Expensive


Cannot detect humans, animals


Susceptible to interference


Front facing standard optical camera







Cheap


Many uses


Simple to install


Begun by Intel, currently maintained by
community, under stewardship of Willow
Garage



Extensive library of Computer Vision
functions



C, C++, Python, Java



No need to continually ‘reinvent the wheel’

Capture
Image

Process
(OpenCV)

Feedback
To User

Capture
Image

Process
(OpenCV)

Feedback
To User

Threshold
Image

Warp
Perspective

Determine
Distance


Remove road surface and highlight objects


Sample road surface in front of vehicle


Remove pixels
±
35 of sampled value


Apply binary threshold

Threshold
Image

Warp
Perspective

Determine
Distance


Distance in image does not change linearly as
vehicle changes position


Inverse Perspective Mapping



Geometric transform which allows us to
remove perspective effect


Threshold
Image

Warp
Perspective

Determine
Distance


All road pixels are zero



Analyze area in front of car



Find first non zero pixels



Translate to distance using scaling factor


How can we know this ‘scaling factor’?



Need to calibrate for particular camera setup



Can be done once for given environment and
parameters


Lay 1m object on road surface


Use chessboard pattern of known size



Roughly calculated for project testing

Capture
Image

Process
(OpenCV)

Feedback
To User


Provide graphical feedback to user


1.
Threshold to remove road surface.

Generate transformation matrix


2.
Transform image to IPM view


3.
Distance determination


4.
Graphics overlay


5.
Modify algorithm for use on a real time video
stream



Further work possible


Improve
thresholding

for different road conditions


Improve performance of IPM algorithm


Automatic calibration implementation



Paper submitted to ISSC 2010, awaiting review


S. Tuohy, D. O
Cualain
, M.
Glavin
, E. Jones:

Distance
Determination for an Automobile Environment using
Inverse Perspective Mapping in OpenCV




Successful implementation of proposed
algorithm

Demonstration