Abstract:
The objective of this project is to develop dehazing, object recognition and lane
recognition algorithms. This report explores different techniques used for the dehazing,
recognition of traffic signals/traffic signs and lane recognition to improve driving
experience in hazy or hazy weather. Different stages involving image processing
occurred in the project many methods and architectures are tested to drive good results.
Module by module implementation occurred in the project the first priority and need
is to dehaze the image so that the dehazed/cleared image for its further processing to
the traffic signals/sign and lane recognition. Haze is the leading cause of car accidents
all around the world. Haze is defined as a moisture droplet in the air that hinders
visibility. Light scatters between droplets when it hits haze, creating a thick white
backdrop. As the number of droplets grows, the haze thickens, making it impossible
for a driver to see his surroundings. Because dense haze distorts the light, drivers
misunderstand the distance between other cars, traffic signs visibility and traffic signal
statuses believing one thing to be far away while it is actually near to them. Driving a
car in this situation is exceedingly unsafe. This study tackles the problem and offers a
remedy by dehazing the haze in real time with Traffic Signals/Signs Recognition and
Lane Assist. Dehazed real-time video can improve the experience ofdriving by aiding
with lane changes and detecting traffic signs and signals. This project uses the
Computer visions, deep learning and DCP (Dark Channel Prior) algorithm to develop
the program. The main advantage ofusing this program is that it enhances the driving
experience by assisting the driver in hazy weather condition. The program works in
step by step prioritized way from dehazing the real-time input to recognition oftraffic
signals/signs and then lane assist.
