Available via license: CC BY
Content may be subject to copyright.
International Journal of Innovative Research in Engineering and Management (IJIREM)
ISSN (Online): 2350-0557, Volume-10, Issue-3, June 2023
https://doi.org/10.55524/ijirem.2023.10.3.19
Article ID IJIR2413, Pages 131-133
www.ijirem.org
Innovative Research Publication 131
LiDAR for Object Detection in Self Driving Cars
Nisha Charaya
Assistant Professor, Department Electronics and Communication Engineering, Amity University, Gurgaon, Haryana, India
Correspondence should be addressed to Nisha Charaya; charayanisha.1010@gmail.com
Copyright © 2023 Made Nisha Charaya. This is an open-access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ABSTRACT- Self driving cars are the major invention in
vehicular automation. These cars use sensors to perceive the
surrounding and control accordingly. Object detection
becomes a major task in these driverless cars. In this study,
utilization of LiDAR to automatically control speed,
braking, and safety systems in response to sudden changes
in traffic conditions is presented. The aim of this study is to
do precise and quick object detection for LiDAR based self-
driving cars. This work proposes image processing to be
used to identify and differentiate similar looking objects so
that carefully calculated driving decisions can be taken.
KEYWORDS- Autonomous Self-driving, Driverless,
LiDAR, Object classification, Object detection
I. INTRODUCTION
A self-driving car is capable of travelling without human
intervention. It is sometimes referred to as a robotic car,
driverless car, or autonomous car. These vehicles sense
their environment using sensors. Control systems analyze
sensory data to build a three-dimensional representation of
the environment around the vehicle. The vehicle then
determines a suitable navigation route based on the model
as well as approaches for navigating traffic restrictions
(such as stop signs) and barriers [1] [2].
These autonomous vehicles are expected to have an impact
on the auto business, as well as the health, welfare, urban
planning, traffic, insurance, labor market, and other
industries once the technology is more
developed. However, it may face few technological
obstacles as under:
In tumultuous inner-city settings, artificial intelligence
is still unable to perform as intended.
A car's computer could potentially be compromised, as
could a communication system between cars.
The ability to avoid large animals requires recognition
and tracking, and Volvo found that software designed
for caribou, deer, and elk was ineffective with
kangaroos. The susceptibility of the car's sensing and
navigation systems to different types of weather (such as
snow) or deliberate interference, including jamming and
spoofing. They would need to be able to revert to
sensible behaviors in situations where these maps might
be outdated.
Competition for the desired radio spectrum for
communications in cars.
Field programmability for the systems will require
careful evaluation of product development and the
component supply chain.
For automated cars to operate at their best, the current
road infrastructure may need to be modified.
Validation challenge of Automated Driving and need for
novel simulation-based approaches comprising digital
twins and agent-based traffic simulation [2] [3].
In addition to these, precise object detection becomes a
crucial task for relying on these cars.
There are many ways to gather the information from
surrounding to detect an object such as such as optical
and thermo-graphic cameras, radar, LiDAR, ultrasound/
sonar, GPS, odometry and inertial measurement units.
Among these LiDAR is a promising and precise
technology that can be utilized for object detection.
II. TRACKING METHODS
For autonomous driving to be accurate and effective, object
tracking is crucial. The identification of things from
photographs and vehicle sensor data, such as pedestrians,
autos, and other obstacles, is a crucial and challenging
interdisciplinary field. It incorporates contributions from
machine learning, signal processing, and/or computer
vision. The majority of processed sensor data comes in the
form of point clouds, pictures, or a combination of the two.
There are several ways to manage point cloud data, but the
most popular one is some kind of 3D grid where a voxel
engine is used to navigate the point space. When numerous
forms of sensor data are available, registration, point
matching, and image/point cloud fusion may be necessary.
The need to take into consideration temporal cues and
estimate motion from time-based frames makes this task
more challenging [4] [5].
Rarely does a single target appear in the scenes involved in
autonomous driving scenarios. The majority of the time,
several items must be simultaneously detected and tracked,
some of which may be moving with respect to the vehicle
and to other objects. As such, most approaches in the
related literature handle more than one object and are
therefore aimed at solving multiple object tracking
problems (MOT) [5].
A sequence of sensor data is available from one or multiple
vehicle-mounted acquisitions devices. Most related
methods involve assigning an ID or identifying a response
for all objects detected within a frame, and then attempting
to match the IDs across subsequent frames. Given that the
monitored objects may enter and exit the frame at various
timestamps, this is frequently a challenging process. They
might also be blocked from view by their surroundings or
even by one another. Additional problems may be caused
International Journal of Innovative Research in Engineering & Management (IJIREM)
Innovative Research Publication 132
by defects in the acquired images: noise, sampling or
compression artifacts, aliasing, or acquisition errors [4] [5].
The most prevalent need for automatic driving object
monitoring is real-time video. Thus, in addition to
individual object recognition, the goal is to link monitored
objects across numerous video frames. Additional
difficulties arise when accounting for variations in motion,
such as when objects are subject to rotation or scaling
transformations or when their relative movement speeds are
fast [5].
Images are typically the primary mode for interpreting the
scene. As a result, 2D MOT is the focus of many efforts in
the associated literature. These techniques rely on a series
of detection and tracking phases, in which successive
detections that have the same classification are connected to
establish trajectories. The inherent existence of noise in the
captured photos poses a substantial difficulty because it
may negatively alter the attributes of identical objects
across consecutive frames. Therefore, computing robust
features is a crucial component of object detection. Color,
frequency and distribution, shape, geometry, contours, or
relationships within segmented objects are just a few
examples of the many different object qualities that features
might represent. The most widely used feature detection
techniques nowadays use supervised learning. Machine
learning algorithms are used to gradually refine features,
which initially start off as collections of random numbers.
Such methods call for the careful selection of hyper-
parameters and the use of appropriate training data,
frequently discovered through trial and error. The best
results, however, in terms of accuracy and robustness to
affine transformations, occlusion, and noise are provided by
supervised classification and regression approaches,
according to a number of findings from the associated
literature [5].
III. PROPOSED METHOD
LiDAR combined with image processing can give precise
results for object detection. Further it can be combined with
speed-distance measurements so that the control system can
take decisions accordingly [6].
The term "light detection and ranging," or LiDAR, is
frequently used to refer to a type of sonar that employs
pulsed laser beams to measure the distance to nearby
objects. As LiDAR systems have shrunk in size, more uses
have emerged that make use of the technology's
adaptability, accuracy, and record-breakingly quick data
collection. Most notably, carmakers are leveraging LiDAR
capabilities as a key component in their race to develop
safe, self-driving vehicles [7] [8].
Both ADAS (Advanced Driver Assistance Systems) and
autonomous vehicles use LiDAR because its sensors enable
accurate, trustworthy navigation during real-time
autonomous operation on highways and in urban areas. In
order to help vehicles safely move at varied speeds, they
can identify and track other vehicles, pedestrians, and other
objects. This includes traveling night and day in a range of
road conditions such as rain, sleet, and snow. However, it
can’t differentiate between a plastic bag and an obstacle; a
cyclist giving hand signal to change lane or a rod projecting
out from a pillar and many more which may affect the
control of autonomous cars. This raises the need for a
precise object detection based on inputs obtained from
LiDAR sensors so that similar appearing objects can be
differentiated [9] [10].
To overcome these issues, object detection techniques of
image processing can be combined with LiDAR based
autonomous cars in such a way that different cloud points
obtained from these sensors are clustered together to form a
recognizable shape as shown in figure 1.
Figure 1: Clustering of Cloud Points [10]
Thereafter, image classification techniques are used to
identify these shapes as objects as shown in figure 2.
Figure 2: Classification of Shapes formed [11]
Thereafter modeling of classified objects is done to predict
all possible movements with speed as shown in figure 3.
Figure 3: Modeling of Classified Objects [11]
International Journal of Innovative Research in Engineering & Management (IJIREM)
Innovative Research Publication 133
IV. OBJECT CLASSIFICATION
Finding objects in point cloud data is the initial step in
recognizing and classifying them using a variety of
techniques and algorithms. After converting raw data into a
point cloud structure, one of the first steps is the point
clustering or segmentation, which basically consists in
grouping points based on common characteristics. After this
step, redundant data can be removed from the point cloud,
resulting in less data to be transferred and processed in the
upcoming phases. Some methods begin by categorizing the
point cloud into background and foreground data in
applications where the sensor maintains a stationary
position. Because they don't depict dynamic objects, points
that are in the same place in several frames are disregarded
as backdrop. For the remaining points (foreground), the
distance between points is measured, and points close to
each other are clustered and marked with a bounding box as
they possibly represent an object. However, when the
sensor moves with the car, these approaches are not
effective, as the background and objects move together
inside the point cloud. Therefore, automotive approaches
require robust and faster algorithms since the objects in the
point cloud also change at higher frequencies. Initial
approaches used sliding windows algorithms with Support
Vector Machine (SVM) classifiers and hand-crafted
features for object detection, but were quickly replaced by
newer superior techniques such as 2D representations,
volumetric-based, and raw point-based data, which deploy
machine learning techniques in the perception system of the
vehicle .These classification /object detection results along
with LiDAR based distance and speed measurements can
help in a more accurate control of self-driving
cars[12][13][14].
V. CONCLUSION
Autonomous cars are a big milestone in automation of
vehicles. Cars with advanced driver assistance that provide
numerous features such as cruise control are already there
in market which aids in comfortable driving. However,
switching to autonomous cars is still far ahead because of
its technological challenges. LiDAR combined with suitable
algorithm for object classification technique can be a
potential approach for object detection.
REFERENCES
[1] S. I. N. Y. H. D. A. N. S. karkra, N. charya, “Unmanned
Ground Vehicle”, IJRITCC, vol. 5, no. 5, pp. 773–775, May
2017.
[2] Xie, S.; Hu, J.; Bhowmick, P.; Ding, Z.; Arvin, F.,
"Distributed Motion Planning for Safe Autonomous Vehicle
Overtaking via Artificial Potential Field" IEEE Transactions
on Intelligent Transportation Systems, 2022.
[3] ^ Gehrig, Stefan K.; Stein, Fridtjof J. (1999). Dead
reckoning and cartography using stereo vision for an
automated car. IEEE/RSJ International Conference on
Intelligent Robots and Systems. Vol. 3. Kyongju. pp. 1507–
1512. doi:10.1109/IROS.1999.811692. ISBN 0-7803-5184-
3.
[4] https://www.labellerr.com/blog/object-tracking-in-
autonomous-vehicles-how-it-works/
[5] https://encyclopedia.pub/entry/8270
[6] Alam Bhuiyan, Ifte Khairul. (2017). LiDAR Sensor for
Autonomous Vehicle. 10.13140/RG.2.2.16982.34887/1.
[7] Shahian-Jahromi Babak et al 2017 IOP Conf. Ser.: Mater.
Sci. Eng. 224 012029
[8] https://www.automotiveworld.com/articles/LiDARs-for-self-
driving-vehicles-a-technological-arms-race/
[9] https://www.geospatialworld.net/prime/business-and-
industry-trends/LiDAR-the-driving-technology-for-
autonomous-and-semi-autonomous-mobility/
[10] https://www.csail.mit.edu/news/more-efficient-LiDAR-
sensing-self-driving-cars
[11] https://www.ff.com/us/futuresight/what-is-LiDAR/
[12] Arnold, E.; Al-Jarrah, O.Y.; Dianati, M.; Fallah, S.; Oxtoby,
D.; Mouzakitis, A. A Survey on 3D Object Detection
Methods for Autonomous Driving Applications. IEEE
Trans. Intell. Transp. Syst. 2019, 20, 3782–3795.
[13] Shi, S.; Wang, X.; Li, H. PointRCNN: 3D Object Proposal
Generation and Detection from Point Cloud. In Proceedings
of the 2019 IEEE/CVF Conference on Computer Vision and
Pattern Recognition (CVPR), Long Beach, CA, USA, 16–20
June 2019; pp. 770–779.
[14] Wu, J.; Xu, H.; Tian, Y.; Pi, R.; Yue, R. Vehicle Detection
under Adverse Weather from Roadside LiDAR Data.
Sensors 2020, 20, 3433.
ABOUT THE AUTHOR
Dr. Nisha Charaya is working as an
Assistant Professor in Department of
Electronics and Communication
Engineering, Amity University Haryana.
Her research areas include signal
processing, image processing, biometrics
and machine learning. She has over 13
years of experience in academics and has
authored numerous research papers in
conferences and journals.