Smart Intelligent Computing and Communication Technology 35

V.D. Ambeth Kumar et al. (Eds.)

© 2021 The authors and IOS Press.
This article is published online with Open Access by IOS Press and distributed under the terms
of the Creative Commons Attribution Non-Commercial License 4.0 (CC BY-NC 4.0).
doi:10.3233/APC210009

Lane Detection and Tracking Algorithm

Based on Curve Fitting Model
Rajakumar.Ra,1, Pandian.Rb, and PremJacob.Tc,
a
UG Scholar, Dept of ECE, Sathyabama Institute of Science and Technology, Chennai,
India
b
UG Scholar, Dept of ECE, Sathyabama Institute of Science and Technology, Chennai,
India
c
UG Scholar, Dept of ECE, Anna University, MIT Campus, Chennai, India

Abstract. Lane Detection is the fundamental vehicle driving system and self-
driving. The proposed concept is to employ the pixel difference in the expected lane
and the road backdrop to detach the lane from the road, and then the Curve fitting
model is made use of in the segregated lanes to locate the straight line in the image
as the lane line. This paper offers a lane detection method based on the Sobel filter
and Curve-fitting model for lane line tracking in different conditions. The main
objective is to improve the accuracy of the Xi’an city database and the KITTI vision
benchmark suite dataset. To achieve this HLS color space was performed which
identifies the lane by adding pixel values.

Keywords. Sobel filter, Curve fitting model, Lane detection, Lane tracking, sliding
window, deep neural network.

1. Introduction

Most accidents occur due to invisible road lanes. The accidents can be reduced
drastically, by employing improved driving assists. A system that warns the driver, can
save considerable number of lives..Indriving assistance to achieve safety on roads, the
challenging tasks are road lanes’ detection or boundaries’ detection which is exposed in
white and yellow lines on roads. In this work, we developed a Curve fitting model
improving the robustness of detecting and tracking the lane for safe transportation.In our
method, inspection of lane and its tracking will be done by the Curve fitting model and
related component function to improvise the working of the lane detection and tracking.

2. Related Works

The literature[1] extracted the Adaptive Region of Interest to minimise computational

complexity. Then Kalman filter is used to find road boundaries sensed in the AROIusing
Progressive Probabilistic Hough Transform (PPHT) in the succeeding frame.
1
Rajakumar.R,UG Scholar, Dept of ECE, Sathyabama Institute of Science and Technology, Chennai, India.
e-mail:[email protected]
36 Rajakumar.R et al. / Lane Detection and Tracking Algorithm Based on Curve Fitting Model

In literature [2] lane marks in road images are extracted which is based on the multi-
constraints model and a clustering algorithm based is proposed to detect the lane. The
literature [3] used the B-spline fitting lane line from the strength of the RANSAC
algorithm for detection offront-view images and improved Hough transform to find
rear-view images. The literature [4] improved the accuracy of lane recognition and
reduced the difference between pixels. The literature [6] proposed an flexible road
detection technique that connects both lane lines and boundaries of obstacle ,
applicable for detecting lanes. The literature [7] makes use of a color threshold method
to identify lane edges with perspective transformations and Hough transform to find
out lane segments in the image with measured conditions. These conditions comprise a
straight road and a sunny weather. In literature [8] based on the Vision-Based
Extrapolation method which performs well in controlled weather conditions and Hough
transform used to detect only the straight lane. In this paper just as in [5] edge-based
detection with the open street map is used to detect lanes which increases computation
time. Hough Transform [9] is employed to identify the straight lane and the Curve
fitting model detects the curve lane which increases the computation time [13-14].

3. Lane Detection Algorithm

Lane detection in images is a significant operation in image processing and computer

vision. Lane detection algorithms detect the lane edges and determine the vehicle
position in the lane. A single monocular camera is used for lane detection. This work
corresponds to the accurate location of the vehicle in the same lane. This system
recognizes the majority of white and yellow markings across the lane effectively during
different conditions which include shadows, rain, snow, or other disturbances on the
road. The preprocessed image is perspective transformed which converts a 3D image to
a 2Dimage.Then the Sobel filter was performed for noise reduction and to identify the
pixel representing the edge. The sliding window method is applied from the bottom of
the image by identifying lane pixels. To calculate the vehicle’s center offset, the
polynomial fit in pixels is used to determine the x position of the left and right lane
corresponding to the y at the bottom of the image.

3.1 Sobel Edge Detection

By applying the threshold a particular part of the image, ROI, we have an HLS color
image as input. In this step, to locate boundaries of lane ,one edge detection method
called the Sobel filter, is used and the boundaries are detected. In Sobel filter, the main
objective is to detect the edges that are nearer to the real lane edges. Sobel edge
detection basically uses the gradient vector of an intense image. The captured color
image is transformed to HLS color space to make the method quicker, computationally
less intensive, and less sensitive to scene conditions. To detect the white lane, lightness
is set to a value close to 100%. Then the combination of saturation and lightness value
was defined to detect the yellow lane. In our proposed method, captured images choose
from the directory of the KITTI Benchmark suite, would be processed. The camera is so
calibrated that the vanishing point of the road should be kept on the top of the Region of
Interest(ROI).
 Rajakumar.R et al. / Lane Detection and Tracking Algorithm Based on Curve Fitting Model 37

Figure 1.Sobel filter and Perspective transform

3.2 Lane Detection and Tracking Flow Chart

Figure 2. Lane detection and tracking flow chart

4. Lane Tracking algorithm

Lane Tracking is mainly employed to increase the computation efficiency of the lane
detection algorithm by keeping the earlier information of how the states have an
estimateofthefuturestates.Thisalgorithmincludesapredictionstepas well as a measurement
step. In the case of Lane Tracking, the prediction stage involves shifting the detected
lanes by a specific amount in the image, based on polynomial fit. In the measurement
step, the radius of curvature and vehicle offset were computed.

4.1 Sliding Window

To draw multiple sliding windows, the starting point of the windows should be known.
To find the initial point, a histogram for the bottom part of the image is calculated.
Based on the peak value of the histogram, the initial window is selected and the mean of
the non-zero points inside the window is determined. For the first half of the image, the
left lane peak is obtained and the other right half gives the peak of the right lane. Thus,
left and right starting sliding windows are formed, and then left lane center and right
lane center are calculated. This kind of selection works fine for both lanes are in the left
38 Rajakumar.R et al. / Lane Detection and Tracking Algorithm Based on Curve Fitting Model

lane on the left side of the image and the right lane on the right side of the image.The
sliding window output is in figure 3.

4.2 Lane Design Parameters

Figure 3.Sliding window output

The equation of the curve model is given as,

Ax2+Bx+C=0 -------equation (1)

As A, B, C are the constants of the quadratic curve, five of which three constants of
the quadratic curve are thus state radius of the curvature is calculated using the y axis
point at the bottom of the image.

R = (1+ (d2y/dx2))3/2 (dy/ dx) 2dx, dy are the lane positions.

5. Experimental Results

5.1 Lane Detection For Video Frames From Xi’an Citydataset

The present section details the experimental out comes of our method of detection of
lane with two sets of various video frames obtained from the Xi’an city dataset.

Figure4.a,b,c,d Output frame of Xi’an city database

5.2 Radius Calculation Of Xi’an City Database

The radius and vehicle offset ofa lane in the Xi’an city video database calculated using
the equation 2 formula and shown in table 1.
 Rajakumar.R et al. / Lane Detection and Tracking Algorithm Based on Curve Fitting Model 39

Table 1. Radius and Vehicle offset calculation of Xi’an city database

IMAGE RADIUS (METER) VEHICLE OFFSET (METER)
Fig.5.a 4.1007 -0.263
Fig.5.b 4.3386 -0.001
Fig.5.c 6.6480 0.0834
Fig.5.d 6.6102 -0.2037

6. Conclusion and Future Work

Lane detection was done using the HLS color space and edge detection by using the
Sobel filter. Then curve fitting model was used for efficient lane detection and tracking.
For two video frames accuracy is calculated as 94.80% and for the KITTI dataset (not
described inside this paper) accuracy was calculated as 95.83%.This algorithm
computation time was calculated as (150-200milliseconds/frame) which is suitable for
real-time applications. This algorithm can be further developed for self-driving
vehicles.

References

[1] MehrezMarzougui, Areejalasiry, Yassinkortli and JamelbailI ,A Lane Tracking Method Based on
Progressive Probabilistic Hough Transform, IEEE access vol.8 pp. 84893 - 84905, 13 May 2020.
[2] HanyuXuan, Hongzhe Liu1, Jiazheng Yuan, And Qing Li, Robust Lane-Mark
ExtractionforAutonomousDrivingUnderComplexRealConditions,IEEEAcess,Vol. 6, pp.5749-5766,
March 9, 2018.
[3] HuiXiong;DamengYu;JinxinLiu;HeyeHuang;QingXu;JianqiangWang;Keqiang Li, Fast and robust
approaches for lane detection using multi-camera fusion incomplex scenes, IET Intelligent Transport
Systems, Volume: 14, Issue: 12, 12 2020), pp. 1582– 1593, 19 November 2020.
[4] Xingyao Wang, Da Yan, Ke Chen, Yancong Deng, Cheng Long, Kunlin Zhang, Sibo Yan, Lane
Extraction and Quality Evaluation: A Hough Transform Based Approach, 2020 IEEE Conference on
Multimedia Information Processing andRetrieval (MIPR), September 03, 2020.
[5] Guanqi Ding; Xiansheng Qin, Vision-based Edge-lined Assisted Moving Vehicle Detection, 2020
IEEE International Conference on Mechatronics and Automation (ICMA), 13-16 Oct. 2020.
[6] Jing LiJunzheng Wang; Min Yan, Adaptive road detection method combining lane line and obstacle
boundary, IET Image Processing, Volume: 14, Issue: 10, pp. 2216 – 2226, 15 October 2020.
[7] StevanStević, Marko Dragojević, MomčiloKrunić, NenadČetić, Vision-Based Extrapolation of Road
Lane Lines in Controlled Conditions, 2020 Zooming Innovation in Consumer Technologies
Conference (ZINC), August 15, 2020.
[8] Xiaoliang Wang; YeqiangQian; Chunxiang Wang; Ming Yang,Map-Enhanced Ego-Lane Detection
in the Missing Feature Scenarios, IEEE Acess, vol.8, pp. 107958 - 107968, 8 June 2020.
[9] Huifeng Wang, Yunfei Wang, Xiangmo Zhao, Guiping Wang, He Huang, Jiajia Zhang , Lane
Detection of Curving Road for Structural High-way with Straight-curve Model on Vision , IEEE
Transactions on Vehicular Technology vol.68 issue:6 pp. 5321-5330,26 April 2019.
[10] Lim.K.H, et al, Lane detection and Kalman-based linear parabolic lane tracking, International
Conference on Intelligent Human-Machine Systems and Cybernetics .2009. pp.351-354.
[11] Deng.J, and Han .Y. A real-time system of lane detection and tracking based on optimized
RANSAC B-spline fitting. Research in Adaptive and Convergent System, 2013, pp. 157-164.
[12] Gopalan .R, Shneier.M, et al, A learning approach toward detection and tracking of lane markings.
IEEE Trans on International Transportation System, vol. 13,no. 3,pp. 1088-1098, Mar. 2012.
[13] Ambeth Kumar.V.D ,.Malathi.S, Ashok Kumar.V.D (2015) .Performance Improvement Using an
Automation System for Segmentation of Multiple Parametric Features Based on Human Footprint .
for the Journal of Electrical Engineering & Technology (JEET) , vol. 10, no. 4, pp.1815-1821
.2015. [http://dx.doi.org/10.5370/JEET.2015.10.4.1815]
40 Rajakumar.R et al. / Lane Detection and Tracking Algorithm Based on Curve Fitting Model

[14] V.D.Ambeth Kumar et.al,, “An Efficient Security System for Data base Management from Illegal
Access, IEEE International Conference on Wireless Communications, Signal Processing and
Networking, 2016. DOI: 10.1109/WiSPNET.2016.7566252