The interface of the autonomous car with the surroundings must be similar to that of human way of interaction. Humans use their eyes as a source of vision and then processes the visual signals in his/her brain and takes the necessary action accordingly. Similarly the autonomous car uses a camera as a visual source to know its surrounding, path etc. and uses the image processing techniques on the images received from the camera. This processing takes place on a minicomputer (RASPBERRY PI). After image processing, control instructions are passed on to the driving motors which helps in steering the vehicle accordingly.
In the future, automated systems will help to avoid accidents and reduce congestion. The future vehicles will be capable of determining the best route and warn each other about the conditions ahead. Many companies and institutions are working together in countless projects in order to implement the intelligent vehicles and transportation networks of the future.
2. Introduction
Aim of the project is to design a
prototype of self- driving
vehicle(SDV).
SDV can navigate using image
processing on Raspberry Pi
minicomputer.
The two main tasks of our model car
is to detect and navigate through
lanes on the road and detect object.
3. Problem Definition
Now-a-days the number of vehicles on the road are
increasing rapidly leading to large number of accidents.
Most of these accidents occur due to human errors like
distracted (texting, phone calls) and aggressive (drunk
and drive, over speeding) driving.
Therefore to bring down the accident count and ensure
safety of passengers we have to decrease human
involvement and automate the process of driving.
This can be achieved by designing an SELF DRIVING
VEHICLE.
4. Phases of Project
PHASE 1:
To learn image processing techniques and implement
them to detect lanes.
PHASE 2:
To interface camera, motor driver IC with raspberry pi and
implement lane detection and sign detection, navigate
through arena accordingly.
8. Finding lines in an image
Original Image Edge detected Image
Detected Hough Lines
9. Methodology of lane detection
Crop
region of
interest
Image
Processing
Binary Edge
Detected
Image
Hough
lines
Mid point
of Lane
Steer Bot
accordingly
14. Finding the mid point of the lane
From centre check the first red
pixel in the image to left. This
gives the nearest line from the
centre to the left.
From centre check the first red
pixel in the image to right. This
gives the nearest line from the
centre to the right.
15. Bot Steering
If midpoint of the lane is in the left half of the image, we
steer the bot to right.
If midpoint of the lane is in the right half of the image, we
steer the bot to left.
Else straight.
18. Object Detection
Object detection refers to a method of identifying and finding
the object of certain class through out the image
There are many classifiers and machine learning tools available
for object detection, we are using Haar cascade technique which
has high detection rate.
Haar cascade
Haar-cascade system finds the object in the image by a moving
window across the image.
The classifier contains a list of cascade stages.
19. Haar Classifier Working
If classifier labelling gives a negative result, therefore the required
object is not found at that specific region and hereby the window
location is shifted to respective next location.
If classifier labelling gives a result of positive, therefore the object is
found in the image and the classifier is shifted to next stage.
If the classifier labelling gives a verdict of result positive, only when all
stages finds the required object in the image.
20. Steps in Haar training
Collection of
Positive and
negative
images
Create vec
file
Train haar
classifier
Create xml