ROAD LANE DETECTION

SYSTEM USING
COMPUTER VISION
Roopa Chavan
Shridhar Nagathan Tasleem Kumthe
Dept of Computer Science and
Dept of Computer Science and Dept of Computer Science and
Engineering
Engineering Engineering
V P Dr.P.G. Halakatti College Of
V P Dr.P.G. Halakatti College Of V P Dr.P.G. Halakatti College Of
Engineering & Technology
Engineering & Technology Engineering & Technology
Vijayapur, India
Vijayapur, India Vijayapur, India
rcroopali@gmail.com
email address email address
Samreen Bani
Dept of Computer Science and
Engineering
V P Dr.P.G. Halakatti College Of
Engineering & Technology
Vijayapur, India
email address

Abstract— Today’s technological advancement solutions that currently exist only use a single
brings an opportunity to automate all the essential front- facing camera to capture input images,
processes. Driving is one of the challenging tasks. which compromises both system performance and
In recent years, the number of car owners has robustness.
been increasing day by day, and cars equipped Index Terms— AlexNet, VGGNet, ReLU ,
cuDNN v3, CamVid
with self-driving functions are attracting
increasing attention. The approach used in this
study converts images captured from a video into
I. INTRODUCTION
a series of frames, generates image features for
each frame, and uses it further to detect lane
boundaries present on the road. Use.The lane Lane detection is a crucial task. Accurate
adjustment process generates a matrix of variables detection of lane markings on the road is essential
to achieve effective lane detection in changing for vehicle localization, path planning, and ensuring
slope scenarios. This improves the real-time safe driving. Convolutional Neural Networks
performance of lane detection. The trace is (CNNs), with their powerful image recognition
extracted through a pair of hyperbolas that fit the capabilities, have emerged as a leading approach for
edges of the trace using the Hough transform. tackling this problem. This project aims to develop
Convolutional neural network-based approaches a robust lane detection system using CNN,
typically identify the presence of classes within a implemented as a Streamlit application that
given image. Examples of this principle include processes video inputs and live camera feeds to
large-scale neural networks such as AlexNet and identify and mark lane boundaries in real- time. To
VGGNet. The first half of the network performs use B-splines, you need to extract a set of candidate
convolution, ReLU, and downsampling of the points from road markings. One of them is
input image. In the second half, upsampling and particularly important. Lane boundary position is
deconvolution are performed. SegNet was able to necessary to avoid collisions with other vehicles
process his 360x480 images in 422.5 ms on and to locate the vehicle on the road. Some of them
NVIDIA Titian GPU with cuDNN v3 and use LiDAR data to take advantage of differences in
achieved his 60.10 mean union (mIoU) across all the reflectance of road markings. However, LiDAR
classes in the CamVid dataset. Most of the is very expensive and is not always installed in
vehicles. On the other hand, cameras are 2022. Many research have been carried out to
inexpensive and can take advantage of differences minimize accidents as there is huge increase in the
in the color of the road surface. A slightly different numbers, and lack of driver’s attention becomes one
approach is used when using CNNs to estimate of the major cause. Lane detection using IOT has
polyline points and solve the fragmentation gained the attention which works by detecting
problem that primarily occurs in segmentation borders on the road and then prompting the turning
networks. Our system meets the real-time of the road. Road Edge Detection System Using
requirements of NVIDIA Titan XP GPUs. Code Surround View Camera Images, Ashok Dahal ,Eric
Golab ,Rajendra Garlapati, 2021. Lane Detection and
and pre-built models for inference are available.
Tracking in Challenging Scenarios[4], Vinh Truong
Convolutional neural network based approaches
Hoang, Kiet Tran-Trung and Assaye Belay, 2020. An
typically identify the presence of a class in a intelligent transportation system has an important
particular image. Large neural networks such as component known as lane-detection system. We
AlexNet or VGGNet are examples of this build a robust lane-detection-and-tracking algorithm
principle. The first half of the network performs to deal with challenging scenarios such as a lane
convolution, ReLU, and down samples the input curvature, worn lane markings, lane changes, and
image. The second half performs upsampling and emerging, ending, merging, and splitting lane. CNN
deconvolution. SegNet can process a 360 x 480 based lane detection with instance segmentation in
image in 422.5 ms on NVIDIA Titian GPU with edge- cloud computing [5] Wei Wang, Hui Lin and
cuDNN v3, achieving a mean intersection over Junshu Wang 2020. As vehicle owners are
union (mIoU) score of 60.10 on all classes in the increasing, the cars with self-driving functions have
CamVid dataset. Most of them use only a single gained more attention. Combination of cloud
front camera to capture the input image; hence, the computing with lane detection can effectively solve
system’s performance and robustness suffer. They the problems of traditional lane detection such as
have also used traditional computer vision feature extraction and high definition, but there is
algorithms that have recently been shown to have large amount of calculation required. At the same
far inferior performance compared to deep time, cloud data processing combined with edge
computing can effectively reduce the computing load
learning-based systems.
of the central nodes. The traditional lane detection
method is improved, and the current popular
convolutional neural network (CNN) is used to build
II RELATED WORK a dual model based on instance segmentation. Lane
Detection on Roads using Computer Vision [6],
Analysis of Road Lane Detection Using Computer Abhishek Goyal, Mridula Singh, Anand Srivastava
Vision[1], Aditi Pandey Sri Ramswaroop Memorial 2019 Lane Detection is a computer vision task that
College of Engineering and Management 2023 . involves identifying the boundaries of driving lanes
Lane Detection is a computer vision task that in a video or image of a road scene. The goal is to
involves identifying the boundaries of driving lanes accurately locate and track the lane markings in real-
in a video or image of a road scene. The goal is to time, even in challenging conditions such as poor
accurately locate and track the lane markings in lighting, glare, or complex road layouts. A Computer
real-time, even in challenging conditions such as Vision based Lane Detection [7], Md. Rezwanul
poor lighting, glare, or complex road layouts. Lane Haque, Md. Milon Islam, Kazi Saeed Alam, Hasib
detection on roads using computer vision[2], Dr. C Iqbal 2019. Lane detection is considered as an
K Gomathy, Mr. D. Sravan Kumar, Mr. Manoj interesting issue for the advancement of Advanced
Kumar, Mr.Ch Venkata Mohith, 2021, lane Driver Assistance Systems (ADAS). But still, now it
detection is a computer vision task that is a most challenging problem because of some
involves identifying the boundaries of driving lanes factors that are faced by lane detection systems like
in a video or image of a road scene. The goal is to as vagueness of lane patterns, perspective
accurately locate and track the lane markings in consequence, low visibility of the lane lines,
real-time, even in challenging conditions such as shadows, incomplete occlusions, brightness and light
poor lighting, glare, or complex road layouts using .
reflection A Study on Real-Time Detection Method
RNN detectors. Computer Vision and the IoT- of Lane and Vehicle for Lane Change Assistant
Based Intelligent Road Lane[3] R. Shashidhar , B.N. System using Vison System on HighWay[8]
Arunakumari , A.S. Manjunath, Neelu Jyoti Ahuja, VanQuang Nguyen,Heungsuk Kim, SeoChangJun ,
Kwangsuck Boo, 2018. In this work we come up are road lanes or not. Post-processing: The final
with a method to identify the lane for drivers’ output is a map of the road lanes detected in the
assistance. However, the combination of lane and image. Image Resizing: Resize input images to a
vehicle information supports the driver assistance consistent size, reducing computational complexity.
system, or the lane change assistant system to Color space Conversion: Convert images to a
improve the reliability of results. For the lane suitable color space (e.g., grayscale or HLS) to
change assistant system (LCAS), it must detect the highlight lane markings. Noise Reduction: Apply
frontal lanes and discover the vehicles around a test filters (e.g., Gaussian or median) to reduce noise and
vehicle Lane detection and Classification using smooth the image. Contrast Enhancement: Adjust
Cascaded CNNs [8] Fabio Pizzati, Marco Allodi, image contrast to make lane markings more
Alejandro Barrera 2017. Lane detection is extremely distinguishable. Region of Interest (ROI) Selection:
important for autonomous vehicles. Therefore Focus on relevant parts of the image where lanes are
almost all researchers make usage lane boundary expected, ignoring unnecessary details. Edge
information to locate the vehicle inside the street, or
Detection: Use edge detection techniques (e.g.,
to implement GPS based localization. Computer Canny edge detector) to identify potential lane
vision based tasks like convolutional neural boundaries. Hough Transform: Transform detected
networks (CNNs) represent the state-of-the-art edges to the Hough space to identify lines.
technology to recognize lane bounadries. However, Perspective Transformation: Correct for camera
the position of the lane boundaries with respect to perspective to obtain a top-down view, simplifying
the vehicle may not suffice for a reliable lane detection. Shadow Removal: Address lighting
positioning, as for path planning or localization variations and shadows that can impact lane
information regarding lane types may also be visibility. Histogram Equalization: Enhance the
needed. In this work, we present an end-to-end contrast of images to improve lane feature visibility.
system for lane boundary identification, clustering Hough Transform: In road lane detection, the Hough
and classification, based on two cascaded neural Transform is often used to identify lines in an image.
networks, that runs in real-time. Road Lane Here's how it is typically applied: Edge Detection:
Detection using Convolutional Neural Network [9] Use an edge detection method (e.g., Canny edge
Farjana Farvin, Sowndarya 2022, Road Lane detector) to highlight significant changes in intensity,
detection is necessary in self-driving car which can emphasizing potential lane markings. Hough Space
detect road lane automatically. With the increase in Representation: Convert the edge-detected image
population, people are more interested to buy new into the Hough space, where each point in the space
cars and also subjected to more accidents. Increase represents a line in the original image. The Hough
in the number of vehicles, human errors towards space is often parameterized by slope (m) and
traffic rules and the difficulty to oversee situational
intercept (b) in the case of a straight line. The Hough
dangers by drivers are contributing to the majority Transform is particularly useful for detecting lines in
of accidents on the road. images, making it valuable for identifying lane
markings on roads. It is robust against gaps and
interruptions in the lines and can handle various line
orientations. However, tuning parameters and dealing
III METHODLOGY with false positives/negatives are challenges that
practitioners often address in implementing road lane
Lane detection is a critical component of road safety detection systems.
and autonomous driving systems. Various
algorithms have been developed to accurately detect Line Detection: To detect road lanes using computer
and track lane markings on the road. In this section, vision includes these steps:
we will explore three popular lane detection Convert Image to Grayscale. Use a Gaussian blurs to
algorithms: Hough Transform, Canny Edge minimize noise and make subsequent edge detection
Detection, and Deep Learning-based approaches. more robust. Apply Canny edge detector, to highlight
Pre-processing: The input image is pre-processed to significant changes in intensity. Region of Interest
remove noise and enhance contrast. Hough (ROI) Masking: Define a region of interest, typically
Transform: The pre-processed image is then a trapezoidal shape covering the road area, and mask
transformed using the Hough transform algorithm to out irrelevant parts of the image. Apply the Hough
detect lines in the image. Line Detection: The transform to identify lines in the edge-detected
detected lines are then analyzed to determine if they image. This can help detect straight lines in the polar
 coordinate space. Filter and Extrapolate Lines: Filter points. The algorithm uses a stochastic Hough line
detected lines based on slope to separate left and transform.. Draw lines on the Image or Video:
right lane lines. Then, extrapolate these lines to After identifying the trace lines in the region of
cover the full extent of the lane. Overlay on Original interest using the Hough line transform, we overlay
Image. them with the visual input (video stream/image).

Fig 1: Preprocessing dataset

Fig 2: Road detection using canny edge detection

IV RESULTS V CONCLUSION

Capturing and decoding video file: once you have This project aims to develop a CNN-based lane
video file using a VideoFileClip object and initialize detection system integrated into a Streamlit
input, each video frame is decoded (that is, application, capable of processing both video files
converted to an image sequence).Grayscale and live camera feeds to detect and mark lanes in
conversion of image: Grayscale is primarily used to real-time. By leveraging the strengths of CNNs and
convert digital images. This is digital technology. the user-friendly interface of Streamlit, the system
Video images are in RGB format. RGB (red, blue, will provide an effective solution for enhancing the
green) is converted to grayscale. Minimize noise: safety and efficiency of autonomous driving and
Noise can cause false edges. Hence, be sure to ADAS. Through rigorous testing and iterative
perform image smoothing before proceeding. One improvements, the final product will deliver high
way to remove noise from an image is to convolve accuracy and robustness, making it a valuable tool in
the original image with a mask that performs the field of intelligent transportation systems.
smoothing. A Gaussian filter is used to perform this
process. Gaussian filter is a typical image filter
technique to reduce noise and improve image
characteristics. This blurring technique produces REFERENCES
smoother, more visually appealing images by
reducing high-frequency components and improving [1]Analysis of Road Lane Detection Using
overall image quality. Canny Edge Detector: Computer Vision, Aditi Pandey International
Canny Edge Detector is a detection operator that journal of software computing and testing
uses a multi-step algorithm to detect multiple edges. 2023. Vol 9, issue 1, ISSN 2456-2351
Compute the omnidirectional progression of a [2]Lane detection on roads using computer vision,
blurred image and track the edge strength. Hough Dr. C K Gomathy, Mr. D. Sravan Kumar, Mr.
Line Transform: It is a feature extraction method Manoj Kumar, Mr.Ch Venkata Mohith,
used to find basic geometric objects such as lines International Research journal of engineering
and circles. By converting image space to parameter and technology, Vol 8,issue : 10,Oct 2021
space, shapes can be identified by collecting voting [3]Computer Vision and the IoT- Based
 Intelligent Road Lane, R. Shashidhar , B.N.
Arunakumari , A.S. Manjunath, Neelu Jyoti
Ahuja, Hindawi mathematical problems in
engineering,Sept2022.DOI:10.1155/2022/475
5113
[4]. Lane Detection and Tracking in Challenging
Scenarios, Vinh Truong Hoang, Kiet Tran-
Trung and Assaye Belay, 2020.
[5]CNN based lane detection with instance
segmentation in edge- cloud computing Wei
Wang, Hui Lin and Junshu Wang May 2020
https://doi.org/10.1186/s13677-020-00172
[6]Lane Detection on Roads using Computer
Vision, Abhishek Goyal, Mridula Singh,
Anand Srivastava, International Journal of
Engineering and Advanced Technology
9(1):1200-1205, Oct 2019
DOI:10.35940/ijeat.A94256.109119
[7]A Computer Vision based Lane Detection,
Md. Rezwanul Haque, Md. Milon Islam, Kazi
Saeed Alam, Hasib Iqbal, I.J. Image, Graphics
and Signal Processing, 2019, 3, 27-34
Published Online March 2019 in MECS
(http://www.mecs-press.org/) DOI:
10.5815/ijigsp.2019.03.04
[8]A Study on Real-Time Detection Method of
Lane and Vehicle for Lane Change Assistant
System using Vison System on HighWay
VanQuang Nguyen,Heungsuk Kim,
SeoChangJun , Kwangsuck Boo, Engineering
Science and Technology, an International
Journal, Volume 21, Issue 5, October 2018,
Pages 822-833
[9] Road Lane Detection using Convolutional
Neural Network Farjana Farvin , Sowndarya 2
International Journal of Advanced Research in
Science, Communication and Technology
(IJARSCT) Volume 2, Issue 1, June 2022