Thomas Jones completed a project to develop a low-cost driver warning system using computer vision techniques to detect lanes and vehicles. The project objectives were to create a system that could be installed in any vehicle and provide audible warnings to the driver if they veer outside lane markings or get too close to other cars. Investigations found the concept is feasible but more development is needed, especially using additional sensors, to create a reliable prototype. The system was designed to use a Raspberry Pi computer and camera to process video and detect high priority objects to help prevent accidents through driver assistance.

1. Thomas Jones
Engineering BSc
Project summary
The project investigated the feasibility of a
low cost driver warning system with the goal
to improve vehicle safety as many collisions
are down to driver error.
Project Objectives
• Develop a low cost system capable of
detecting lanes and other vehicles that can
be fitted to any car.
• Give the driver audible warnings when the
vehicle is operating outside the operating
parameters
• Study existing computer vision projects
and systems and investigate which
techniques can be used
• Implement an algorithm to develop the
detection speed and accuracy as more
vehicles are detected
Project Conclusion
Investigations have shown that the system is
feasible but needs more development to yield
a reliable working prototype. Use of other
sensors such as proximity and another
camera would increase detection accuracy
and also enable the system to work better at
night or during bad weather.
Project Supervisor
Abdul Farooq
Introduction
Consumer interest in dash cams has been growing
over the past few years as they ensure any
incidents on the road are captured. This has led to
consumers wanting a system to help prevent
accidents, the only issue being the lack of cheaper
options on the market. Developments in Vehicle
Vision and computing power now mean that the
entry barrier to these systems is greatly reduced.
In this system a Camera would be mounted to the
windscreen and used to capture a video feed of
the road ahead. High priority objects such as lane
markings and other vehicles would be detected
and processed by the computer. The program
would give warnings if the vehicle operates
outside of pre-set parameters, such as distance of
the car in front or distance from the centre of the
lane.
Raspberry Pi
A Raspberry Pi 2 was chosen as the development
device as it is both cheap and powerful.
Detection Methods
The Hypothesis Generation (HG) method allows
the locations of vehicles in an image to be
hypothesized, with the objective being to detect
vehicles as quickly as possible.
Vehicle Safety
In recent years’ developments in computing power
now mean that machine vision is being used in an
increasing number of applications. Object
detection is one of the methods that can be
utilised for driver assistance systems. These
systems are continually being developed by
manufacturers, but require integration at the time
of manufacture and when specified as an extra can
cost a substantial amount. The main issue facing
consumers, especially in developing markets is
access to the technology in a package that can be
implemented easily and on a wide range of
vehicles
Product Justification
Research into existing vehicle vision systems
concluded that one of the most popular systems in
use was manufactured by Mobileye. It has been
integrated in many vehicles, including those
produced by BMW, General Motors and
Mitsubishi. The cost of the Mobileye-560 system is
$849, meaning there is a gap in the low-end
segment of the market and so justification of the
research for this product can be made.
Program Design
This flow chart outlines the basic design of the
program. It shows what is being measured whilst
the vehicle is being driven down the road.
Resources
The total cost of the items required for the project
is £58.97. Using open source software will help to
greatly reduce the cost of developing the system
and purchase price for the end user.
Use of 3D vision for vehicle lane guidance and object detection