Vision-Based Lane Detection System for Navigation: A Comprehensive

Review

Sangapu Sreenivasa Chakravarthi

, Kanakala Sri Harika

and Ayyapasetti Abhinav

Amrita School of Computing, Amrita Vishwa Vidyapeetham, Chennai, India

Keywords:

Lane Detection, Autonomous Driving, ADAS, Computer Vision, Hough Line Transform, Canny Edge

Detection, Lane Markers, Region of Interest, Line Extrapolation.

Abstract:

Accurate lane detection becomes a critical requirement in the ADAS and the fully autonomous driving case.

Therefore, this project documents a robust lane detection pipeline with the use of computer vision techniques

such as image preprocessing, edge detection, and Hough Transform for extracting lane markings in differ-

ent road conditions. These open-source libraries, including NumPy, OpenCV, and Matplotlib, constitute the

system. The frames are extracted from videos in the ﬁrst place, followed by preprocessing steps that include

color conversion, Gaussian Blur, and Canny edge detection. A region of interest reﬁnes the detection, while the

Hough Line Transform results in lane markers. Line extrapolation techniques smooth noisy detections across

frames. Analyzing the pipeline under clear weather, shadows, and slight curvature of the road, it manages to

detect lane boundaries reliably with little computational overhead. Hence, it is quite suitable for real-time use.

Future work will extend the system toward more complex environments, such as sharp curves and intersec-

tions.

1 INTRODUCTION

The automotive industry propels forward towards

autonomous driving while lane detection becomes

critical to safety and control. Lane detection tracks

the boundaries of driving lanes, and it is a generally

required component of lane keeping applications, lane

departure warning as well as adaptive cruise control.

Although it was formerly a product of expensive sen-

sors, which include LIDAR and radar, computer vi-

sion offers an alternate approach through RGB cam-

eras coupled with image processing algorithms to rec-

ognize lane markings.

This project focuses on the design of a robust lane

detection system that can work under different varied

road conditions and deal with problems like change-

able light, weather, or different kinds of lane mark-

ings. In this system, there will be an image prepro-

cessing procedure, enhancing contrast and reducing

noise. The performance will be tested with differ-

ent road settings, for example based on changing the

weather, bends of the road and marking of lanes; the

https://orcid.org/0000-0002-8373-7264

https://orcid.org/0009-0000-4023-8812

https://orcid.org/0009-0009-5762-925X

core performance metrics are correctness, processing

speed and robustness.

Despite restricting itself to normal conditions of

driving, the project still lays the foundation for further

development into more complex applications of lane

detection, such as the development of using machine

learning approaches to handle more complex environ-

ments. The general expectation is high accuracy in the

detection of lanes, real-time performance, and effec-

tive visualization of lanes for driving assistance sys-

tems with the potential eventually to contribute to au-

tonomous driving technologies thus improving road

safety in much cheaper ways.

2 RESEARCH METHODOLOGY

2.1 Review Scope

This is a literature review aimed at surveying and

analyzing relevant lane detection techniques for au-

tonomous vehicle systems. This study examines var-

ious classical and modern approaches, starting from

edge detection techniques, the Hough Transform, re-

gion of interest masking, and color space segmenta-

tion, in addition to emerging techniques such as ma-

Sreenivasa Chakravarthi, S., Sri Harika, K. and Abhinav, A.

Vision-Based Lane Detection System for Navigation: A Comprehensive Review.

DOI: 10.5220/0013586800004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 2, pages 71-79

ISBN: 978-989-758-763-4

chine learning, deep learning, and Generative Adver-

sarial Networks-GANs-based methods in pursuing a

solution to the real-time challenge of lane detection.

It then considered the strengths and weaknesses of

these approaches under different scenarios, like light-

ing, shadow, occlusion, and weather changes, in a

critical review of the conventional methods of lane

detection, including Canny and Sobel edge detectors

and Hough Transform. It emphasized the advances

in the state-of-the-art current research trends like ma-

chine learning, deep learning, and real-time applica-

tions, in order to show that the process of lane detec-

tion had passed through a series of handcrafted fea-

ture extraction methods to a highly adaptive, data-

driven approach. There has been special attention

to the strength of these algorithms in terms of han-

dling occlusions, road surface reﬂections, and worn

lane markings.

In the review, by discussing only these speciﬁed

methodologies the focused analysis is brought to the

readers which will cover the essential techniques of

lane detection for autonomous vehicles. This will al-

low the readers to realize better how the methods de-

veloped have to face the challenges in actual driving

scenarios. Included topics do not relate to the princi-

pal issue, such as object detection and vehicle control,

or other navigation issues.

2.2 Research Questions

RQ 1. What are the primary techniques used in lane

detection for multi-lane roads, and how do they ad-

dress the complexities of lane types and trafﬁc condi-

tions?

RQ 2. What challenges do lane detection algorithms

face on multi-lane roads, particularly regarding envi-

ronmental conditions and lane markings?

RQ 3. How do previous works and approaches inform

current methodologies in lane detection, especially in

dealing with varying lane conditions on highways?

RQ 4. In what ways do dataset characteristics inﬂu-

ence the performance of lane detection algorithms on

multi-lane highways?

RQ 5. How do environmental factors like shadows

and reﬂections affect the accuracy of lane detection

algorithms on highways?

RQ 6. What future directions and innovations in lane

detection technology could enhance the reliability and

accuracy of systems for navigating multi-lane high-

ways, considering the challenges discussed?

3 SURVEY OF RELATED WORK

3.1 Lane Detection Techniques

3.1.1 Edge Detection Methods

The Edge detection is used to detect lane markings in

real environments. Among the most popular edge de-

tection techniques, the Canny and Sobel detectors are

among the best in identifying strong gradients of in-

tensity in pixels. The Canny edge detector follows

a multi-stage algorithm that eliminates noise from

images-to put it simply emphasizes edges-through the

use of two Gaussian ﬁlters, and hysteresis threshold-

ing (M. Zaidi, 2024). The Sobel operator changes

gradient images into magnitude and direction using

3x3 kernels that focus on high spatial frequency re-

gions for lane markings (M. Zaidi, 2024). Some re-

cent work has justiﬁed the more traditional methods

by showing their efﬁcacy under different lighting con-

ditions (W. Chen, 2020). Some edge detection algo-

rithms such as Canny and Sobel have highly facili-

tated the process of determination of lane boundaries

in images. These methods rely on intensity changes to

bring out lane markings. They normally fail in cases

of changing lighting conditions or complex road con-

ditions. As put by (Kumar and Simon, 2015), most

early lane detection algorithms lay their base on these

algorithms and require preprocessing to normalize the

variability between different lighting situations.

3.1.2 Hough Transform for Lane Lines

The Hough Transform is one of the most powerful

techniques for detecting straight lines, making this

technique very useful for the identiﬁcation of lane

markings in images. This method transforms points

from Cartesian coordinates into a parameter space

that allows for the effective recognition of lines even

when noise is present (S Kishor, 2024). However,

they have drawbacks: noise sensitivity and complex-

ity in handling curved lane lines (I. Sang, 2024).

Fast Hough Transform and Progressive Probabilistic

Hough Transform (M. A. Javeed, 2023). (M. Mar-

zougui, 2020) it resolves those limitations hence, it

enhances the real-time speed and accuracy in a chang-

ing driving scenario. Although the Hough Transform

can be used to detect straight or slightly curved lane

lines, the method fails when the curve is more com-

plex or at instances of broken lane markings. Accord-

ing to (M. A. Khan, 2022), the algorithm is usually

applied by manual tuning of its parameters in order

to enhance its invariance in different environments.

However, it works best for partially occluded or faded

INCOFT 2025 - International Conference on Futuristic Technology

lane markings.

3.1.3 Region of Interest Masking

Focusing on speciﬁc regions within the image, where

lane markings are more prominent, is an important as-

pect for improving the detection accuracy as well as

reducing the computational load. Techniques such as

region of interest masking isolate the lower region of

the image, which commonly contains the lanes and

ignore other regions that may add noise (L. Ding,

2020). By using ROI algorithms, processing can be

concentrated on relevant data, which further enhances

the overall efﬁciency of lane detection signiﬁcantly

(H. U. Khan, 2020).Such methods minimize the com-

puting load as they direct resources to road sections

where the lane markings are expected, thus enhancing

the detection accuracy by doing away with irrelevant

parts of the image. (J. Huang, 2021) discussed that

adaptive thresholding and ﬁltering techniques further

improve lane detection in terms of performance as

they preselect regions of interest. In general, ROI

masking enables algorithms to optimize their focus-

ing on the areas of a particular image where accuracy

and processing time are improved signiﬁcantly, par-

ticularly for real-time applications in autonomous ve-

hicles (P. S. Perumal, 2023).

3.1.4 Color Space Segmentation for Lane Lines

Color space segmentation makes use of a set of color

spaces (like RGB, HSV) in improvement to the lane

line detection algorithm since the color for the lane

markings is varied. The HSV color space is suit-

able because the hue-based differentiation of color

between lane markings and the road surface is made

possible and saturation and value components en-

able the elimination of undesirable colors (W. Chen,

2020). The thresholding techniques can reliably iso-

late the lane lines even in changing lighting condi-

tions (M. A. Al Noman, 2022). Using color space seg-

mentation techniques colorspace applied by RGB and

HSV color spaces segment the lane based on the color

properties. Color thresholding distinguishes lane

markings from the road surface using color thresh-

olding. (Y. Almalioglu, 2022) suggested it would be

an appropriate approach in detecting lane lines during

broad daylight, nighttime, etc, but there may be some

inﬂuence by the color of the road surface.

3.2 Challenges in Lane Detection

3.2.1 Shadows and Light Variability

Environmental factors, such as trees and buildings,

create lane markings, while the changes in light affect

the detection. These changes in light cover the lane

markings and produce false positives (M. M. Yusuf,

2020). Advanced algorithms need to distinguish be-

tween shadow and actual lane markings to maintain

appropriate detection with dynamic lighting. For in-

stance, lighting variations, shadows, or bright sun-

light can determine the accuracy of lane detection.

The paper introduced pre-processing requirements as

histogram equalization, aiming to normalize lighting

conditions before lane detection by (Y. Almalioglu,

2022). Shadows from trees or buildings can make al-

gorithm detection difﬁcult since sometimes they ob-

scure the lane markings. Variations in light condi-

tions along a given day affect how visible the lane

is and, thus, have inconsistent detection performance

(A. Pandey, 2023). Hence, techniques that normalize

these effects are intrinsically important for robust lane

detection systems.

3.2.2 Weather Conditions and Road Surface

Reﬂection

Weather conditions such as rainfall, fog, or snow

can signiﬁcantly degrade the visibility level of lanes.

Wet or shiny surfaces can further blur lane mark-

ings, making them difﬁcult to capture accurately

(N. A. Rawashdeh, 2022). Robust lane detection sys-

tems have to include adaptation mechanisms toward

these weather variations through dynamic threshold-

ing techniques and ﬁltering that impact reﬂection.

This includes bad weather, which degrades visibility

and creates poor reﬂection on the surface. (H. Jeon,

2022) stated that such adverse conditions have to be

faced and overcome by designing lane detection sys-

tems. They added that either data augmentation or

adaptive algorithms could be followed for higher ro-

bustness. The bottom line is that these challenges

demand adaptive algorithms with real-time capabil-

ities, which can adapt to environmental conditions

(S. Wang, 2023).

3.2.3 Occlusions by Other Vehicles and Objects

These are caused by vehicles, pedestrians, or road

signs that can also compromise lane detection by

blocking the visibility of lane markings (J. F. Rojas,

2023). Detection algorithms that use contextual infor-

mation and the surrounding vehicle can help resolve

the challenge of predicting lane positions in spite of

Vision-Based Lane Detection System for Navigation: A Comprehensive Review

such occlusions (M. A. Javeed, 2023). The estimation

of lane positions when portions of the lane are oc-

cluded uses predictive algorithms and sensor fusion,

such as LiDAR and camera integration. (J. Huang,

2021) discussed the fusing of multiple sensor data to

improve the robustness of lane detection in occluded

scenes. Overall, the occlusions from both vehicles

and roadside objects tend to degrade the lane detec-

tion accuracy, thus requiring highly sophisticated al-

gorithms that have the capability of detecting lane

boundaries even when they are partially occluded

(S. Anbalagan, 2023).

3.2.4 Lane Line Wear and Occlusion

Degraded or not properly maintained lane marking,

along with the temporary closure of the lane, are some

issues that pose difﬁculties in the identiﬁcation of lane

boundary (A. Pandey, 2023). In these cases, the de-

tection algorithms have to be developed with possibly

machine learning algorithms as the different patterns

may or may not be more evident in lane markings

(M. A. Javeed, 2023). Faded lane markings, not well-

maintained roads, temporary lane closures, and con-

struction are some of the tough problems. Therefore,

robust lane detection algorithms have to adapt to the

varying quality of lane lines using deep learning tech-

niques for better generalization in such cases. Lane

markings are more likely to degrade over time due

to wear and tear and hence become less visible. The

lanes which have faded or worn do not make it easy

to spot and require advanced methods in the identiﬁ-

cation processes of these lines of markings (J. Huang,

2021).

3.3 Previous Work and Approaches in

Lane Detection

3.3.1 Hand-Crafted Features

Early algorithms based their detection on hand-

engineered features to detect lane markings, which

included edge gradients, color information, and inten-

sity contrasts (W. Chen, 2020). Effective, yet rigid in

nature and adapted only to limited boundary condi-

tions these approaches often call for signiﬁcant tuning

in different environments. (M. N. Rahaman, 2021)

emphasized the fact that such methods do not adapt

well. Previous approaches primarily relied on man-

ual selection and engineering features from images

for lane detection purposes. As effective as they were,

the methods could not match the adaptability and gen-

eralization capability of modern learning-based ap-

proaches (A. Pandey, 2023).

3.3.2 Machine Learning-Based Methods

Lane detection has now improved since the emer-

gence of machine learning. Machine learning tech-

niques supplemented classical methods with the in-

vention of making the model learn complex patterns

from labeled datasets, making it increase detections

in trying environments (L. Ding, 2020). The change

thus improves robustness and accuracy under diverse

conditions (N. J. Zakaria, 2023). The development

of machine learning methodologies, rather than rely-

ing only on handcrafted feature extraction, allowed

the models to learn from data, and this consequently

resulted in better accuracy in detection tasks in dif-

ferent settings. Thus, adaptive models that can even

recognize complex patterns replaced the handcrafted

feature extraction. (A. Mehra, 2020) reported some

of the early machine learning-based methodologies

using Support Vector Machines and Decision Trees.

Recent years have witnessed further developments

in machine learning methods that utilize large-scale

datasets to train models capable of lane detection un-

der diverse scenarios. These developments enhance

the robustness and detection accuracy vastly in com-

parison with traditional approaches (M. H. Saikat,

2024).

3.3.3 Deep Learning Approaches

Recent works in lane detection techniques employ

several machine learning (ML) and deep learning

(DL) models to enhance the autonomous vehicle per-

formance under different scenarios. Ding et al. pre-

sented a CNN-based approach focusing on the se-

mantics of segmentation that could successfully ex-

tract lane features well. Along a similar trend,

(M. M. Yusuf, 2020). built up a fully convolu-

tional network, FCN that adopts data augmentation

methodologies to enhance robustness towards ad-

verse weather and illumination conditions. Moti-

vated by efﬁciency requirements, (M. A. Khan, 2022)

have proposed a lightweight model called LLDNet

to achieve real-time lane detection with surprisingly

modest performance gains at low computational re-

sources. (J. Huang, 2021) utilized SegNet coupled

with LiDAR data in the context of real-time detec-

tion of lane and curbs, while (H. Jeon, 2022) have

used U-Net in a simulation environment to demon-

strate the feasibility of a lane detection event in heavy

rain conditions. (W. Chen, 2020) considered re-

current neural networks (RNNs) for temporal lane

marking tracking and have emphasized the impor-

tance of sequential data processing in a lane de-

tection approach.(Y. Zhang, 2020) presented Ripple-

GAN, which is a GAN that improves both qual-

INCOFT 2025 - International Conference on Futuristic Technology

ity and the robustness of the generated lane line

images, thus enhancing the detection performance.

(G. Li, 2022) studies deep reinforcement learning

strategies for decision-making in lane change scenar-

ios, thereby indirectly affecting the detection capabil-

ity. (A. Pandey, 2023) added attention mechanisms

to further enhance the performance. Other recent

researches by (S Kishor, 2024) proposed the fusion

of Canny edge detection with CNNs for lane detec-

tion and (P. S. Perumal, 2023) developed LaneScan-

NET to simultaneously detect lane states and obsta-

cles. (S. Wang, 2023) focused on improving accu-

racy in detection using dynamic region of interest,

and (S. Anbalagan, 2023) developed a vision-based

lane departure warning system. Apart from that, sev-

eral works are worthy of mention. For example, there

is (M. A. Al Noman, 2022), which applies gradi-

ent threshold techniques in the case of lane detection

under varying weather and other conditions. Then,

(S. Sultana, 2023) proposed robust detection in chal-

lenging environments. Lastly,(L. Caltagirone, 2018)

investigated the application of LIDAR-camera fusion

with fully convolutional networks for enhanced road

detection, and (J. Q. Zhang, 2023) proposed a model

called HoughLaneNet based on the integrated deep

Hough transform and dynamic convolution to im-

prove the efﬁciency of lane detection. Taken together,

these works further exemplify how lane detection al-

gorithms are developed, addressing underlying chal-

lenges and making autonomous vehicles more reli-

able in various circumstances of driving.

3.3.4 Innovative Applications of GANs for Lane

Detection

GANs have been implemented in lane detection in

many innovative ways. Some of the latest and notable

work includes the unpaired image-to-image transla-

tion method CycleGAN, which was used for trans-

lating images from clear to adverse weather condi-

tions to facilitate the training of robust models for lane

detection in diversiﬁed scenarios. Similarly, while

the application of Conditional Generative Adversarial

Networks in medical imaging is mainly demonstrated

for the generation of synthetic lane images for serving

as supplements to the training datasets for lane detec-

tion systems. More successes have been demonstrated

using Ripple-GAN proposed by (Y. Zhang, 2020).

In that, a customized network is used together with

a Wasserstein GAN which enhances the quality and

robustness of the lane line images produced. Over-

all, these studies point toward the direction by which

GANs can revolutionize lane detection, either by re-

producing challenging scenes that are hard to collect

under real-world conditions or by augmenting avail-

able training datasets.

3.3.5 Real-Time Lane Detection with

Autonomous Vehicles

The necessity of lane detection in real-time in an au-

tonomous vehicle puts pressure on the development

of fast vision processing systems. According to Jo-

han Fanas Rojas et al., as far as reliable detection is

concerned, there are myriad sensors connected to ad-

vanced algorithms that present dependable real-time

lane detection for safe use of an autonomous vehicle.

In fact, for an autonomous vehicle, real-time lane de-

tection is critical. Higher-order systems utilize multi-

task learning and sensor fusion to ensure detection

accuracy with high computational efﬁciency. Other

techniques included in the autonomous driving frame-

work are reinforcement learning and end-to-end deep

learning to track proper lanes even in challenging en-

vironments. As concluded by (V. Mus

at, 2021) who

said ”combining the data from various sensors, such

as cameras, LiDAR, and radar, improves lane detec-

tion performance and reliability in real-time applica-

tions” . Real-time lane detection is very critical to the

safe navigation and operation of autonomous vehi-

cles. Recent works have developed image-processing

algorithms that can be able to work in real time and

hence guarantee correct lanes’ detection and interpre-

tations as the vehicle moves (P. S. Perumal, 2023). It

shows that speed and reliability are amongst the most

critical factors in success in the lane detection sys-

tems.

3.4 Dataset

3.4.1 Environment Type

Another highly comprehensive urban dataset is

KITTI, which contains information for city streets

and highways. CULane captures various urban roads

and highways and consists of lane structures and chal-

lenging scenarios such as lane changes and occlu-

sions. ApolloScape, another dataset, includes high-

density urban and highway environments with high

lane markings and annotations of the scene. Lane

detection for highways In cases with lesser clutter,

the highway-speciﬁc lane detection is covered by

the clean and simple annotations of lane markers in

TuSimple dataset. (P. S. Perumal, 2023) develops for

the urban driving scenario; it can use the CULane

dataset, while (S. Anbalagan, 2023) discuss the lane

departure warning system, which most probably is

based on an urban dataset as KITTI or ApolloScape.

Further, (M. A. Al Noman, 2022) gives techniques

for lane detection suitable for rural areas by utilizing

Vision-Based Lane Detection System for Navigation: A Comprehensive Review

datasets such as KITTI that are known to cover a wide

range of scenarios.

3.4.2 Sensor Modalities

The lane detection datasets most used are camera-

based, such as CULane (X. Pan, 2018), TuSimple

(S. Yoo, 2020), and is based on 2D vision lane

detection, while BDD100K (F. Yu, 2018) captures

videos from dashboard cameras for both front-facing

and surrounding views. However, KITTI dataset

(A. Geiger, 2017) combines LiDAR and GPS data for

integrated sensor analysis; and the nuScenes dataset

takes (H. Caesar, 2020) this further with 6 cameras,

5 radars, and 1 LiDAR, thus becoming ideal for re-

search in multimodal sensor fusion for lane and object

detection.

3.4.3 Weather and Lighting Conditions

These datasets are remarkable due to various kinds of

weather conditions such as rain, fog, and night-time

driving, which create really challenging scenarios for

the detection algorithms. Still, ApolloScape focuses

on high-resolution lane detection in diverse weather

conditions. Culane is targeted for speciﬁc night-time

driving and poor visibility conditions, which makes

useful for even testing the robustness of lane detec-

tion algorithms, further complemented by BDD100K

including night and dusk scenes.

3.4.4 Complexity of Road and Lane Markings

As for TuSimple (S. Yoo, 2020), since it utilizes high-

way driving with obvious and simple markings, it can

be used for the preliminary testing of deep learning

models on lane detection; however, the road condi-

tions in CULane (X. Pan, 2018) and ApolloScape

(X. Huang and Yang, 2018) are more complex, with

heavy trafﬁc, occlusion, and densely populated lanes,

which are challenging conditions to the lane detection

models especially in the urban driving where mark-

ers are faded and often changed, while pedestrians or

other obstacles appear.

3.4.5 Dataset Scale

Among the largest available datasets, BDD100K

(F. Yu, 2018) comes with 100,000 video clips that

may be appropriate for training deep models that con-

sume large amounts of data; CULane (X. Pan, 2018),

on the other hand, contains more than 133,000 la-

beled frames. Medium-scale models and resources

like KITTI (A. Geiger, 2017) contain around 15,000

frames, and TuSimple (S. Yoo, 2020) with 6,408

training video clips which may be highly applicable

to test a new lane detection method without requiring

tremendous computational power.

3.4.6 Challenges and Special Features

The nuScenes dataset (H. Caesar, 2020) is apt for sen-

sor fusion research since it uses information coming

from multiple cameras, several radars, and LiDAR

sensors to improve lane detection under adverse con-

ditions. ApolloScape (X. Huang and Yang, 2018), on

the other hand, is targeted at high-deﬁnition map cre-

ation and scene parsing, with rich annotations for lane

markings, road boundaries, and trafﬁc signs. Hence,

this is aptly suited for lane detection in environments

that have complex geometries.

4 SUMMARY OF THE REVIEW

FINDINGS

As the survey comes to a close, the research

questions can now be addressed using the insights

gained from the review.

Response to RQ1 Lane detection techniques for

multi-lane highways utilizes a variety of advanced

methods to tackle the complexities of diverse lane

types and trafﬁc conditions. Deep learning archi-

tectures, particularly convolutional neural networks

(CNNs), effectively extract intricate patterns from

highway imagery, achieving high accuracy in lane

marking detection. Ensemble approaches, such as

those by (M. A. Khan, 2022), enhance robustness

by integrating multiple models, and improving

detection during lane changes and occlusions. Hybrid

methodologies combining traditional techniques

like the Hough Transform with CNNs, as shown

by (S Kishor, 2024), maintain accuracy in complex

scenarios. Preprocessing techniques, such as region

of interest (ROI) masking, optimize computational

resources by focusing on relevant image areas.

Together, these methods enhance the reliability and

safety of autonomous vehicles in dynamic driving

environments, ensuring accurate lane detection even

under challenging conditions.

Response to RQ2 Lane detection algorithms

on multi-lane highways encounter numerous chal-

lenges that stem from complex environmental

conditions and the nature of lane markings. Envi-

ronmental factors such as shadows cast by overhead

structures and ﬂuctuations in natural light can ob-

scure lane visibility, complicating accurate detection

(M. M. Yusuf, 2020). Additionally, adverse weather

INCOFT 2025 - International Conference on Futuristic Technology

conditions—ranging from rain and fog to snow—can

degrade the clarity of lane markings, making them

harder to detect (N. A. Rawashdeh, 2022).

Multi-lane highways present unique difﬁculties,

as lane markings can become worn or faded over

time, further diminishing their visibility (A. Pandey,

2023). Moreover, the presence of occlusions caused

by other vehicles and roadside objects can signiﬁ-

cantly hinder the ability to accurately discern lane

boundaries (J. F. Rojas, 2023). Advanced algorithms,

such as those employing sensor fusion techniques,

are crucial for maintaining robust lane detection

performance despite these challenges (J. Huang,

2021). By leveraging a combination of contextual

information and sophisticated processing methods,

researchers are continually developing strategies to

enhance detection reliability on multi-lane highways,

thus ensuring safer navigation for autonomous vehi-

cles in diverse driving conditions.

Response to RQ3 Current lane detection

methodologies for highways are deeply inﬂuenced

by earlier works that established foundational tech-

niques. Traditional edge detection methods, like

Canny and Sobel, remain crucial for identifying

lane markings through intensity changes, as shown

by (M. Zaidi, 2024). The evolution to machine

learning has enhanced accuracy by enabling models

to learn complex patterns from labeled datasets

(L. Ding, 2020). Region of interest (ROI) masking

techniques have improved detection efﬁciency by

focusing on relevant areas (M. A. Khan, 2022).

Deep learning, particularly CNNs, has achieved

remarkable performance in challenging conditions,

as demonstrated by (M. M. Yusuf, 2020). Innovations

such as sensor fusion and GANs have emerged from

these foundational studies, improving robustness in

adverse weather scenarios (S. Wang, 2023). Together,

these advancements illustrate how past approaches

inform current methodologies in lane detection on

highways.

Response to RQ4 The performance of lane

detection algorithms on multi-lane highways is

intricately linked to the characteristics of the datasets

utilized for training and evaluation. Diverse datasets,

such as the CULane and ApolloScape, offer varied

environmental conditions and lane conﬁgurations,

enabling algorithms to learn and adapt effectively.

For instance, CULane’s rich annotations of complex

urban scenarios enhance the algorithm’s robustness

against occlusions and dynamic lane changes, as

noted by (X. Pan, 2018). Additionally, the integration

of sensor modalities in datasets like nuScenes allows

for the fusion of information from multiple sensors,

which signiﬁcantly improves detection accuracy

under adverse conditions, as emphasized by (H. Cae-

sar, 2020). Moreover, the scale of datasets, such

as the extensive BDD100K with its 100,000 video

clips, provides ample training data that helps deep

learning models generalize better to various trafﬁc

scenarios, thereby addressing challenges associated

with lane markings’ variability, as highlighted by

(F. Yu, 2018). Overall, the combination of diverse

environments, multimodal sensors, and large-scale

data sets is crucial in enhancing the performance of

lane detection algorithms on multi-lane highways.

Response to RQ5 The accuracy of lane de-

tection algorithms on highways is signiﬁcantly

inﬂuenced by environmental factors such as shad-

ows and reﬂections. Advanced algorithms must

adeptly differentiate between true lane markings and

the artifacts caused by shadows or surface reﬂec-

tions, which can obscure visibility. For instance,

(M. M. Yusuf, 2020) highlight that variations in

lighting, particularly due to shadows cast by trees or

overpasses, can lead to false positives in lane marking

detection. Moreover, the paper by (Y. Almalioglu,

2022) emphasizes the importance of pre-processing

techniques, like histogram equalization, to normalize

lighting conditions prior to lane detection, thereby

mitigating these challenges. Without these corrective

measures, detection accuracy can suffer, as incon-

sistent lighting conditions often lead to unreliable

results, particularly during peak sunlight hours or

adverse weather situations (A. Pandey, 2023). Thus,

incorporating adaptive algorithms capable of real-

time adjustments is essential for robust lane detection

performance in dynamic highway environments.

Response to RQ6 Future directions and in-

novations in lane detection technology could focus on

several key areas to enhance reliability and accuracy

for navigating multi-lane highways. Advances

in machine learning, particularly deep learning

techniques, can facilitate the development of more

robust algorithms capable of adapting to diverse

lane conditions and environmental challenges. In-

corporating multi-modal sensor fusion, combining

data from cameras, LiDAR, and radar, could signif-

icantly improve detection performance in complex

scenarios. Additionally, utilizing synthetic data

generation through methods like GANs can aug-

ment training datasets, enabling algorithms to learn

from a wider range of conditions, including rare or

difﬁcult-to-capture scenarios. Furthermore, ongoing

research into real-time processing capabilities and

Vision-Based Lane Detection System for Navigation: A Comprehensive Review

the integration of emerging technologies, such as

V2X (Vehicle-to-Everything) communication, could

provide valuable contextual information to enhance

lane detection systems. Ultimately, a focus on these

innovations will be crucial in developing reliable lane

detection solutions that can navigate the complexities

of multi-lane highways effectively.

5 CONCLUSION

This review outlines the evolution of lane-

detection techniques with a focus on the revolution-

ary leap through sophisticated methodologies such

as deep learning and traditional algorithms. State-

of-the-art applications that allow integrating CNNs

for remarkable effectiveness in accurately detecting

lane markings across diverse environmental condi-

tions and varied trafﬁc scenarios are increasing. Hy-

brid approaches that combine classical methods, such

as the Hough Transform, with modern deep-learning

techniques greatly increase robustness in issues like

occlusions, variable lighting, and faded markings. In

addition, preprocessing techniques like masking the

ROI optimized computations by focusing the process-

ing power on areas of interest and hence improving

the accuracy of detection in real-time applications.

Despite these advances, problems related to environ-

mental variability and degradation of lane marking

have yet to materialize. Research into sensor fusion,

adaptive algorithms, and robust preprocessing meth-

ods can ensure lane detection systems in autonomous

vehicles are dependable and safe, thus leading to ef-

fective and secure road guidance.

6 FUTURE SCOPE

The algorithm to be proposed is based on the thor-

ough literature survey that is conducted, aiming to

overcome the current limitations in detecting lanes

under different conditions. The algorithm, applies a

sequential approach, starting with the preprocessing

steps including converting to grayscale and applying

a Gaussian blur to maximize the image quality with

less noise. Edge detection is carried out by the Canny

method and followed by a region of interest (ROI)

masking to ﬁlter out critical roadway features. Hough

Line Transform is then applied to detect lane bound-

aries using linear patterns. After post-processing, it

further reﬁnes and gives more accurate lane detec-

tion results. Some further steps have been used to

explore and draw lane boundaries. The ﬁnal pro-

cessed output is then integrated with LaneScanNet,

a deep learning-based model that distinguishes itself

based on its adaptability to various conditions and

also its real-time efﬁcient working scenario. Unlike

the existing approaches, the hybrid pipeline connects

the traditional computer vision techniques with the

deep learning methodologies while ensuring robust

lane detection in even challenging environments. The

proposed methodology will set the stage for future ad-

vancements in autonomous navigation with safety, ac-

curacy, and adaptability to various road and weather

conditions, which existing algorithms cannot achieve.

REFERENCES

A. Geiger, Philip Lenz, C. S. R. U. (2017). Are we ready for

autonomous driving? the kitti vision benchmark suite.

In IEEE Conference on Computer Vision and Pattern

Recognition.

A. Mehra, Murari Mandal, P. N. V. C. (2020). 1reviewnet: A

fast and resource optimizednetwork for enabling safe

autonomous driving inhazy weather conditions.

A. Pandey, R. S. (2023). Analysis of road lane detection us-

ing computer vision. In International Journal of Soft-

ware Computing and Testing.

F. Yu, Wenqi Xian, Y. C. F. L. M. L. W. L. T. Y. T. D. (2018).

Bdd100k: A diverse driving video database with scal-

able annotation tooling.

G. Li, Yifan Yang, S. L. X. Q. N. L. S. E. L. (2022). Deci-

sion making of autonomous vehicles in lane change

scenarios: Deep reinforcement learning approaches

with risk awareness.

H. Caesar, Varun Bankiti, A. H. L. S. V. V. E. L. Q. X. A.

K. Y. P. G. B. O. B. (2020). nuscenes: A multimodal

dataset for autonomous driving. In IEEE Conference

on Computer Vision and Pattern Recognition.

H. Jeon, YaeOhn Kim, M. C. D. P. S. S. J. L. G. C. Y. L.

(2022). Carla simulator-based evaluation framework

development of lane detection accuracy performance

under sensor blockage caused by heavy rain for au-

tonomous vehicle. In IEEE Robotics and Automation

Letters.

H. U. Khan, Afsheen Rafaqat Ali, A. H. A. A. W. K. A. Z.

(2020). Lane detection using lane boundary marker

network with road geometry constraints. In IEEE Win-

ter Conference on Applications of Computer Vision.

I. Sang, W. R. N. (2024). A robust lane detection algorithm

adaptable to challenging weather conditions. In Inter-

national conference on computer vision.

J. F. Rojas, Pritesh Patil, A. M. M. T. H. B. A. R. E. Z. D. A.

(2023). Automated vehicle lane centering system re-

quirements informed by resilience engineering and a

solution using infrastructure-based sensors.

J. Huang, Pallab K. Choudhury, S. Y. L. Z. (2021). Real-

time road curb and lane detection for autonomous

driving using lidar point clouds. In Banan Science

and Technology Foundation of Chongqing, China.

INCOFT 2025 - International Conference on Futuristic Technology

J. Q. Zhang, Hao-Bin Duan, J.-L. C. A. S. M. W. (2023).

Houghlanenet: Lane detection with deep hough trans-

form and dynamic convolution.

Kumar, A. M. and Simon, P. (2015). Review of lane detec-

tion and tracking algorithms in advanced driver assis-

tance system. In International Journal of Computer

Science & Information Technology.

L. Caltagirone, Mauro Bellone, L. S. M. W. (2018). Lidar-

camera fusion for road detection using fully convolu-

tional neural networks.

L. Ding, Huyin Zhang, J. X. C. S. S. L. (2020). A lane detec-

tion method based on semantic segmentation. In In-

ternational Conference on Intelligent Robots and Sys-

tems.

M. A. Al Noman, Zhai Li, F. H. A. M. F. R. B. O. S. R.

S. M. C. W. (2022). A computer vision-based lane

detection technique using gradient threshold and hue-

lightness-saturation value for an autonomous vehicle.

In International Journal of Electrical and Computer

Engineering.

M. A. Javeed, Muhammad Arslan Ghaffar, M. A. A. N. Z.

A. S. M. M. E. M. T.-E. P. B. M. S. J. X. J. (2023).

Lane line detection and object scene segmentation us-

ing otsu thresholding and the fast hough transform for

intelligent vehicles in complex road conditions. In

Advanced Mechatronics and Robotics Technologies in

Industry 4.0 Era: Intelligence and Automation.

M. A. Khan, Md Foysal Haque, K. R. H. S. H. A. M. B. M.

M. A.-A. N. (2022). Lldnet: A lightweight lane detec-

tion approach for autonomous cars using deep learn-

ing. In IEEE Conference on Computer Vision and Pat-

tern Recognition.

M. H. Saikat, Sonjoy Paul Avi, K. T. I. T. T. M. S. A. T. I.

(2024). Real-time vehicle and lane detection using

modiﬁed overfeat cnn: A comprehensive study on ro-

bustness and performance in autonomous driving.

M. M. Yusuf, Tajbia Karim, A. F. M. S. S. (2020). A ro-

bust method for lane detection under adverse weather

and illumination conditions using convolutional neu-

ral network. In ICCA 2020: International Conference

on Computing Advancements.

M. Marzougui, Areej Alasiry, Y. K. J. B. (2020). A lane

tracking method based on progressive probabilistic

hough transform. In IEEE 10th Conference on Sys-

tems, Process & Control.

M. N. Rahaman, Md Shovon Biswas, S. C. M. M. H. S. A.

M. B. (2021). Lane detection for autonomous vehicle

management : Pht approach. In 24th International

Conference on Computer and Information Technol-

ogy.

M. Zaidi, Hamza Daud, M. S. H. A. J. (2024). Lane detec-

tion in autonomous driving: A comprehensive survey

of methods and performance. In IEEE 1st Karachi

Section Humanitarian Technology Conference.

N. A. Rawashdeh, Jeremy P. Bos, N. J. A.-A. (2022). Cam-

era–lidar sensor fusion for drivable area detection in

winter weather using convolutional neural networks.

In Photo-Optical Instrumentation Engineers.

N. J. Zakaria, Mohd Ibrahim Shapiai, R. A. G.-M. N. M. Y.

M. Z. I. N. W. (2023). Lane detection in autonomous

vehicles: A systematic review.

P. S. Perumal, Yong Wang, M. S. S. T.-S. B. M. K. S. V. U.

K. R. (2023). Lanescannet: A deep-learning approach

for simultaneous detection of obstacle-lane states for

autonomous driving systems. In IEEE Conference on

Computer Vision and Pattern Recognition.

S. Anbalagan, Ponnada Srividya, B. T. S. G. S.-G. S. G. R.

(2023). Vision-based ingenious lane departure warn-

ing system for autonomous vehicles.

S Kishor, Rekha Nair, T. B. S. S. (2024). Lane detection

for autonomous vehicles with canny edge detection

and general ﬁlter convolutional neural network. In In-

ternational Conference on Computing for Sustainable

Global Development.

S. Sultana, Boshir Ahmed, M. P. M. R. I.-S. A. (2023).

Vision-based robust lane detection and tracking in

challenging conditions.

S. Wang, Yu Wang, Y. L. (2023). Robust lane detection

method based on dynamic region of interest. In IEEE

Sensors Journal.

S. Yoo, Hee Seok Lee, H. M. S. Y.-H. P. J. C. D. H. K.

(2020). End-to-end lane marker detection via row-

wise classiﬁcation.

V. Mus

at, Ivan Fursa, P. N. F. C. A. B. (2021). Multi-

weather city: Adverse weather stacking for au-

tonomous driving.

W. Chen, Weixing Wang, K. W. Z. L. H. L. S. L. (2020).

Lane departure warning systems and lane line detec-

tion methods based on image processing and seman-

tic segmentation: A review. In Journal of Trafﬁc and

Transportation Engineering.

X. Huang, Xinjing Cheng, Q. G. B. C. D. Z. P. W. Y. L.

and Yang, R. (2018). The apolloscape dataset for au-

tonomous driving.

X. Pan, Jianping Shi, P. L. X. W. X. T. (2018). Spatial

as deep: Spatial cnn for trafﬁc scene understanding.

In Proceedings of the European Conference on Com-

puter Vision.

Y. Almalioglu, Mehmet Turan, N. T. A. M. (2022). Deep

learning-based robust positioning for all-weather au-

tonomous driving. In Conference on Robot Learning.

Y. Zhang, Zongqing Lu, D. M. J.-H. X. Q. L. (2020).

Ripple-gan: Lane line detection with ripple lane line

detection network and wasserstein gan.

Vision-Based Lane Detection System for Navigation: A Comprehensive Review