Obstacle Avoidance with Machine Vision

Obstacle Avoidance using
Machine Vision
Joose Rautemaa
A control system for a car that can avoid obstacles using
Machine Vision
Developed on and for a model car, could possibly be used
in a real car too with minor modifications
Hardware includes a Raspberry Pi and a Raspberry Pi
Remote control capabilities over WLAN
Autonomous navigation will be based on a digital compass
and GPS waypoints
Relatively safe and fast autonomous travel
Recognizing obstacles with the camera allows avoiding them or
going around them if possible
Using OpenCV for the recognition of objects
Ability to easily select the destination and also
manually remotely control the car
A simple web interface that can be used with a laptop or a mobile
phone over WLAN where a user can enter coordinates, select
coordinates from a map, or manually control the car
The interface has an option for a direct camera feed
The Raspberry Pi is a very small and inexpensive device, but it is
not hugely powerful. For faster object detection you would
need a more powerful computer
The Raspberry Pi is very energy efficient which allows for much
longer operating times if using electric power only
OpenCV is really reliable and efficient in recognizing objects in
video feeds, however it requires a lot of processing power
GPS navigation might have to be on a separate device, for
example on an Arduino that is attached via a serial link, because
the Rasbperry Pi might not have enough prosessing power to
handle that too.
OpenCV documentation
Adafruit Learning Community
Various blogs and websites
Various forums and IRC channels
Peers and professionals
What next?
Need to find a good model car that can be used as a
base for actual testing of the control features
Need to start designing the GPS navigation system
and the web interface
The latest version of OpenCV needs to be compiled
for Raspberry Pi, this process might take up to 10