ADAS & DMS

Literature Review

Supervisors: Presented By:

Dr. Jahnavi Yarlagadda A. SRUTHI (SE23MTRE001)
Dr. Ravi Sekhar

1
Content
• What is ADAS

• ADAS Features & Functions

• What is DMS

• DMS Features & Functions

• Types of DMS Alerts

• Papers related to ADAS & DMS

• References

2
ADAS
What is ADAS?
 Advanced driver-assistance systems (ADAS) are technologies that assist drivers with the
safe operation of a vehicle.
 They help in preventing accidents by providing real-time alerts and automated responses.

3
ADAS Features
• Adaptive Cruise Control (ACC)
• Lane keeping assistance system (LKAS)
• Automatic emergency braking (AEB)
• Blind Spot Detection (BSD)
• Traffic Sign Recognition (TSR)
• Parking Assistance system

4
ADAS Features
Adaptive Cruise Control (ACC):
 With the help of adaptive cruise control, an innovative driver assistance technology,
your car will automatically change its speed to keep a safe distance behind the car in
front of you.

5
ADAS Features

Lane keeping assistance system (LKAS) :

 Uses cameras to detect the position of your lane lines.
 Emits a beeping or vibrates the steering wheel when you begin to drift out of the
lane.
 It gently corrects your steering automatically to move you back into the lane.

6
ADAS Features

Automatic emergency braking (AEB) :

 It start braking automatically if a collision is imminent and the driver is not taking any action.
 Detects a potential collision and activate the braking system to decelerate the vehicle to avoid
collision.

7
ADAS Features

Blind Spot Detection (BSD):

 Sensors on the car's sides detect nearby vehicles in the blind spots and alert the driver.
 This technology helps make lane changes safer and reduces the chance of side collisions.

8
ADAS Features
Traffic Sign Recognition (TSR):
 TSR technology uses cameras to detect and identify traffic signs such as speed limits and
prohibited entrance markers while driving.
 It then displays this information on the dashboard, keeping the driver informed about local
traffic laws, which is particularly useful in unfamiliar areas.

9
ADAS Features

Parking Assistance system:

 Parking assistance technology has the potential to reduce crashes that happen while parking
 It use sensors and cameras to calculate the best way to navigate your car into a parking spot.

10
ADAS Functions

• Forward Collision Warning (FCW)

• Lane Departure Warning (LDW)
• Pedestrian & Cyclist Collision Warning (PCW)
• Headway Monitoring & Warning (HMW)
• Wrong- way Driving Warning
• Tire Pressure Warning
• Electric Vehicle Warning Sounds

11
ADAS Functions

Forward Collision Warning (FCW):

 The FCW system provides sound and visual warnings when a collision with another
vehicle is imminent.
 It alerts the driver up to three seconds before a potential collision, allowing time to react
based on distance and speed calculations.

12
ADAS Functions

Lane Departure Warning (LDW):

 The system sounds a warning and displays a visual alert when there is a lane deviation
without signaling.
 If the driver doesn't respond, the car takes control and makes necessary adjustments to
stay in the lane.

13
ADAS Functions

Pedestrian & Cyclist Collision Warning (PCW):

 The technology alerts the driver visually and audibly if a collision with a cyclist or
pedestrian is imminent, giving them time to react.
 A Pedestrian Collision Warning (PCW) is triggered when the Time to Collision (TTC) is
calculated to be less than two seconds.

14
ADAS Functions

Headway Monitoring & Warning (HMW):

 HMW continuously measures the gap in seconds between itself and the car in front. If the
distance exceeds the fleet's predetermined threshold, an alert is sent out.

15
DMS

What is DMS?
 The driver monitoring system(DMS) makes use of technology to determine whether the
person operating the car is paying attention and staying alert.

16
Workflow of DMS
Input Video

Face Detection Mouth

Detection

Eye Rub
Eye Detection Yawn Detection
Detection

If Drowsy Eye
rubbed Blink Detection Yawn
Eyes count>3
Yes If
No Drowsy Yes
Yes
Driver Alert !!!

17
DMS Function

Drowsiness Detection:
 DMS monitors the driver's eye movements and head motions to spot signs of fatigue or
drowsiness.
 By identifying these signs, the system can help prevent collisions caused by tired drivers.

18
DMS Function

Distraction Monitoring:
 DMS identifies in-car distractions like using a phone, eating, or talking to passengers.
 By spotting these distractions, the system helps keep the driver's attention on the road.

19
Types of alerts
1. Phone usage alert
2. Smoking alert
3. Fatigue alert
4. Distraction alert, drowsiness, and attention
5. Driver abnormal alert
6. Equipment occlusion
7. Driver identification/absence alert
8. Driver Yawn alert
9. Driver sleep alert
10. Cameras block alert
11. Seatbelt alert, and more..

20
 Paper-1

Comprehensive Study of driver behavior systems using

computer vision and machine learning techniques
Authors - Qu et al.,2024 ( Journal of Big Data)

21
Objective

• Develop AI-based monitoring software

to enhance the safety of self-driving
cars.
• Improve overall vehicular safety by
detecting and identifying various
driving habits that might cause
distractions.
• Provide a safer and more secure
transportation environment.

22
Type of data collected

1. State Farm Distracted Driver Detection dataset (from Kaggle):

 Training images: 22,424 annotated images
 Testing images: 79,726 unannotated images
2. Distracted Driver Dataset (developed by researchers at the American University
in Cairo, the Technical University of Munich, and Valeo Egypt):
 Training images: 12,555 annotated images
 Testing images: 1,923 unannotated images

23
 Methodology

Data Algorithm
Training & Continuou
Preprocessin Developmen
Validation s Learning
g t

• Data Preprocessing: Preprocessing is done on the images in the datasets to improve their
quality and fit for AI model training.
• Algorithm Development: The AI approach uses machine learning algorithms, particularly
CNNs, to identify and categorize driving behaviors.
• Training & Validation: The AI model is trained with preprocessed photos and its accuracy in
identifying driving behaviors is verified using annotated photos from both datasets.
• Continuous Learning: The system supports continuous learning, improving over time by
incorporating new data and adjusting its algorithms based on posture data.

24
Results & Conclusions

• The paper discusses the safety issues of self-driving cars and emphasizes the
importance of AI in understanding driver behavior.
• It showcases an advanced AI system that works better than current methods.
• The study explores ways to evaluate driver behavior and predicts future
advancements in safety technology.

25
Paper-2

Applying GIS Analytics and ADAS with DMS to

Assessment Heavy-Duty Vehicles Unsafe Acts and
Unsafe Conditions in Thailand.
Author- N Pumpugsri ,et al.,2023

26
Objective

• In this paper, ADAS with DMS are

used to help analyze driver fatigue on
the Asian Highway.
• It enhance the ability of the authorities
to improve unsafe environment and
form a strategy to manage and reduce
the accidents.

27
Type of data collected

1. Accident Data from the Department of Land Transport (2019-2021):

Collected data on accidents occurring between 2019 and 2021 from the
Department of Land Transport.
2. Accident Data from the Ministry of Transport (2019-2021): Gathered data on
accidents from the Ministry of Transport for the same period (2019-2021).
3. Current Status of Asian Highway Data in Thailand: Acquired information from
the Department of Land Highways regarding the status of the Asian Highway in
Thailand.
4. Near-Miss Data from ADAS with DMS Devices: Collected near-miss data from
246 Heavy-Duty Vehicles (HDVs) equipped with Advanced Driver Assistance
Systems (ADAS) and Driver Monitoring Systems (DMS) during the period of May
to July 2022.

28
Summary

• GIS used to data preparation, segmentation and screening- Accident analysis.

• Shuttle Radar Topography Mission Digital Elevation Model (SRTM DEM) it collects
photos of topographic maps, information on environment and altitudes.
• Structural contours are important for showing elevation on a topographic map.
They help represent the height and shape of the land.
• ADAS- provide warning to the driver for safe operation, DMS- Monitor the driver
condition.
• Hot spot analysis- shows the accident occurred places.

29
Results & Conclusions

• ADAS and DMS can detect risky driving behaviors such as not signaling when
changing lanes and looking down while driving.
• The data from these systems can be used to alert drivers about dangerous areas,
enhancing road safety.
• When researchers combine GIS analysis with ADAS and DMS, they can more
accurately evaluate events by accounting for topographical and geographic
characteristics.

30
Paper-3

An Integrated ADAS for Assessing Risky Situations in

Urban Driving.
Author-TH Weisswange, et al..,2013

31
Objective

• Advanced Driver Assistance System (ADAS) that enhances traditional physical

collision warnings by incorporating implicit predictions to estimate potential risks.
• Warns dangerous situation through Human Machine Interface (HMI).

32
Type of data collected
• The system was tested in real-time across different weather conditions: sunny,
overcast, cloudy, and snowy.
• Testing involved seven experiments over 22 minutes of driving.
• During these tests, 767 cars were encountered, with 67 identified as risky.
• The driving speed ranged from 6 to 11 meters per second.
• 4.5 % False classification shown in HMI and 95.5 % Shows correct.

33
Methodology

34
Methodology

35
36
Results & Conclusions

• This study shows that ADAS controls typical dangerous situations and possible
hazards from stationary cars by combining situation analysis and enhanced
vision modules.
• Improved the identification of vehicles by adding an orientation detection stage
and building a decision tree to assess potentially hazardous situations.
• Testing on a controlled course was successful, but real traffic occlusions can
cause missed warnings for partially hidden vehicles.

37
Paper-4

Driver State Monitoring System- A Review

Author- A Guettas,et al.., 2019

38
Objective
• Major Accidents are happening due to human, so to overcome this and make safe
environment by implementing DMS in vehicles.
• And discuss this technology, talking a little about its development, operation,
standards.

39
Methodology

40
41
Paper-5

Detection of drowsiness using a DMS

Author- C Schwarz,et al.., 2019

42
Objective
• Drowsiness is the major cause of driver impairment so in this paper we study about
high-fidelity driving simulator to evaluate automotive production- ready driver
monitoring system (DMS) to detect drowsy driving.

43
Type of data collected
• 20 licensed adult drivers (50% female, are range 18-66).
• 4hrs drive along 45-mile interstate nighttime loop with 15 min break.
• Drove between 10 p.m. to 2 a.m.
• Visual Data: Facial expressions, eye movements, and head position.
• Physiological Data: Heart rate, blink rate, and other relevant biometrics.

44
Methodology

45
Results & Conclusions
• The Driver Monitoring System (DMS) effectively detects driver drowsiness with high
accuracy, especially when combining visual and physiological data.
• The system issues timely alerts, helping to reduce the risk of accidents caused by
drowsy driving.
• The study successfully created a reliable DMS, with future improvements aimed at
accommodating different drivers and integrating with other safety systems.

46
Paper-6

Optimization Algorithm for Driver Monitoring System using Deep Learning Approach
Author- MW Yoo, DS Han, 2020

47
Objective
• The research aims to create a smart algorithm to improve the Driver Monitoring
System (DMS) using advanced AI techniques.
• The goal is to make the system work faster and more efficiently in real-time, making
it easier to use in cars.
• This improved system will help detect if a driver is sleepy or not paying attention,
making driving safer.

48
Methodology
Driving Start

Enter camera image

Region of interest setting

Gaze and Head pose

calibration

Detection of Gaze and head

pose
Determine the driver’s
condition

DMS optimization
algorithm
49
Results & Conclusions
• The new algorithm accurately and quickly detects if the driver is sleepy or
distracted, and it sends alerts right away.
• The system makes monitoring the driver's head and eye movements simpler and
more reliable.
• This improvement makes driving safer and more efficient, with future plans to detect
other driver activities like using a phone, smoking, or eating.

50
Paper- 7

Impact of Advanced Driver Assistance Systems on

Driving Distraction and Risky Behavior: An Empirical
Analysis of Irish Commercial Drivers
Author- L Masello, et al.., 2023

51
 Objective
• To understand how Advanced Driver Assistance Systems (ADAS) affect driving
distractions and risky behaviors in light commercial vehicle drivers in Ireland.
• It specifically looks at the long-term effects of real-time warnings on improving
driver behavior.
• The study also seeks to identify the factors that make drivers more likely to become
distracted.
• We can also see the evolution of distraction events over time to evaluate whether
the installation of warning-based ADAS helps to improve risky driving behavior.

52
 Data Set
• Data from 373 light commercial vehicles (LCVs) operating in the Republic of
Ireland between 01/05/2020 and 14/03/2022.
• The dataset comprises 16,777,145 km and 988,849 trips, where each driver
drives 135 km daily, on average.
• The dataset contains: driving exposure attributes (total number of trips and
distance travelled) and aggressiveness attributes (total number of harsh
acceleration, harsh braking, and speeding events) also includes distraction
attributes (total number of driver inattention, driver making calls, driver yawning,
driver smoking, forward collision, lane departure warnings).

53
 Methodology
• The GNSS tracker Ruptella ECO 4 light for exposure and aggressiveness
attributes.
• The Streamax C6D AI camera system composed of a road-facing and a driver-
facing camera for the distraction events

54
Methodology

55
Attributes
• Relationship between distraction events and contextual attributes are done using
ARM(Association Rule Mining).

56
Attributes

57
Attributes

58
59
60
Results

• There is a significant decrease in distraction events over time – mainly driver

inattention, forward collision, and lane departure – after 100 days of using
warning-based ADAS.

61
Results

62
Conclusion

• The study found that ADAS, especially with real-time feedback, effectively reduce
distractions and risky behaviors in drivers.
• Using both monitoring systems and training or reward programs can help keep
drivers safer over time.
• Understanding specific situations that increase risk can help improve road safety
by guiding better safety strategies.

63
Questions…
1. Data - Weather condition data, Type of vehicle, Data collected continuous or
discontinuous, time of data collected at daytime or nighttime ,location of the
place, age of the user, Traffic condition(heavy traffic, light traffic).
2. Did they do any driver survey for knowing any information from the driver about
the ADS and DMS functions, how they are interacting with them.
3. About action recognition and activity behavior(head light, signals).

64
Reference
• Qu, F., Dang, N., Furht, B., & Nojoumian, M. (2024). Comprehensive study of driver behavior monitoring
systems using computer vision and machine learning techniques. Journal of Big Data, 11(1), 32.
• Pumpugsri, N., Rattanawong, W., & Vongmanee, V. (2023, September). Applying GIS Analytics and
ADAS with DMS to Assessment Heavy-Duty Vehicles Unsafe Acts and Unsafe Conditions in Thailand.
In 2023 27th International Computer Science and Engineering Conference (ICSEC) (pp. 434-439).
IEEE.
• Weisswange, T. H., Bolder, B., Fritsch, J., Hasler, S., & Goerick, C. (2013, June). An integrated ADAS
for assessing risky situations in urban driving. In 2013 IEEE Intelligent Vehicles Symposium (IV) (pp.
292-297). IEEE.
• Guettas, A., Ayad, S., & Kazar, O. (2019, October). Driver state monitoring system: A review.
In Proceedings of the 4th International Conference on Big Data and Internet of Things (pp. 1-7).
• Schwarz, C., Gaspar, J., Miller, T., & Yousefian, R. (2019). The detection of drowsiness using a driver
monitoring system. Traffic injury prevention, 20(sup1), S157-S161.
• Yoo, M. W., & Han, D. S. (2020, February). Optimization algorithm for driver monitoring system using
deep learning approach. In 2020 International Conference on Artificial Intelligence in Information and
Communication (ICAIIC) (pp. 043-046). IEEE.

65
Paper- 1

Investigation of Bus Drivers’ Reaction to ADAS Warning

System: Application of the Gaussian Mixed Model.
Xiaomeng Shi et al..,

66
Objective
• The main objective is to explore bus driver’s reaction to LDW and FCW with
naturalistic driving data with different types of environment.
• To reduce the road crashes by using ADAS warning for Bus drivers.
• In this we will see how driver respond to LDW & FCW for 20-days naturalistic
data.
• Drivers' reactions are categorized as positive, negative, or emergency using the
Gaussian mixture model.
• A framework was developed to measure and analyze drivers' reactions in various
environments.

67
Methodology

68
69
Data Set
• 20 buses were selected, and all are equipped with Mobile eye M360 System.
• Warning data contain 9 dimensions
1. Environmental information- latitude, longitude, altitude and time.
2. Vehicle running condition- speed and angle.
3. Driver information- age and milege.
4. Alarm duration.

• The GPS data were processed with Map API of AutoNavi, and road types of each
alarm position were obtained.
• Road types classified as three categories: freeways, urban roads, and other road
types (such as ramps and rural roads).

70
Brief about the paper
• According to Birrell, S.A. et al. in 2014, after using FCW the mean headway
increases to 2.3 s and time spent traveling closer than 1.5 s (Dangerous) to the
vehicle in front decreases nearly three times.
• Parameters: Driving information- Videos, GPS, driving characteristics. Personal
information- Gender, age, identity card information, driving license information.
• Driver reaction is different from simulator and real world.
• ADAS Warning may differ from a car driver’s reaction and bus driver’s reaction.
• For data-based driver behavior classification problem, commonly used
unsupervised clustering (such as k-mean), latent Dirichlet (LDA), and support
vector machine(SVM).
• The membership degree is an important concept in fuzzy mathematics that can
be applied in recognition driver behavior and driver assistance system.

71
Limitations
• Long term use of ADAS may reduce driver perception road risk & weaken ability
to address emergencies.
• LDW & FCW cannot control mobility of vehicle.
• Active driver implication and driver don’t have trust on the warning system.
• Driver may lose spontaneity if he knows someone is monitoring.

72
Results
• Driver respond better to FCW than to LDW.
• Drivers react more positively on highways than in urban environment.
• Drivers' reaction were better at night than during the day.

73
Paper-2

Effect of an integrated collision warning system on risk

compensation behavior: An examination under
naturalistic driving conditions
Bo Yu, Shan Bao et. Al.,

74
Objective
• To examine how the introduction of advanced driver-assistance systems (ADAS)
influences drivers' risk compensation behavior, particularly in terms of hard
braking frequency.
• To investigate key factors, such as distraction, that contribute to changes in hard
braking behavior among both teen and adult drivers.

75
Data
• Total 6060 event data in that we have 3030 valid to hard braking and 3030
without hard braking.
• There are 14 input variables, from 6 categories:
1. Integrated warning system: with warning system activated or turned off;
2. Road geometry: road type, lane width, curve and intersection;
3. Traffic environment: traffic density, time of day, and the road surface;
4. Vehicle kinematics: lane change, time headway, and speeding;
5. Driver characteristics: age and gender;
6. Distracted driving: yes or no.

76
Data Selection Criteria

77
Summary
• They have used Honda Accord EX sedans equipped with an integrated collision
warning system.
• Five cameras were mounted in each vehicle to record the driver’s face.
• Time headways categorized into three groups.
1. Aggressive following - (<1s); Lane width - (<3m)
2. Medium following - (1-2s) ); Lane width - (3-5m)
3. Safe following - (>2s) ); Lane width - (>3m)
• Instead of warning systems, distractions (especially visual-manual distraction like
phone use), speeding, and driving in heavy traffic were the main cause of hard
braking.

78
Data analysis
• In this study Random forest is used to find the hard braking behaviour.
• Here, hard braking (<-0.3g) was chosen as dependent variable to find the
occurrence of hard braking behaviour.
• The basic process of Random forest consist of following steps:
• Step 1: Bootstrap sample.
• Step 2: Growing trees.
• Step 3: OOB error estimate.
• Hierarchical logistic regression is used to examine the driver in different age
groups.

79
Results
• Prediction results of Random forests.

80
Results
• Comparison with other machine learning model-
1. Support vector machine learning – 77.1% accuracy.
2. Decision tree – 79.9% accuracy.
3. K-nearest neighbors – 69.3% accuracy.
4. Random Forest – 83.2% accuracy.

81
Results
• Hierarchical logistic regression model results for 95% confidence interval.
• Hard braking behavior lower for younger drivers (-6%), middle aged driver (-38%)
and older drivers (-5%).

82
Limitation
• There are insufficient high- severity crashes(such as fatal or injured crashes)
observed in NDS due to limited exposure periods.
• The 10-second data extraction before hard braking may miss important
information, though it was enough to analyze driving behavior related to factors
like distractions and traffic conditions.
• The braking threshold was lower than in other studies, but the analysis still
improved accuracy and helped understand driver reactions to crash warnings.

83
Paper- 3

Identification of Unskillful Driver Based on Driving

Behavior and Adaptation of Safety and Comfort
Functions for Driver Assistance
Narmada Balaji Robert Bosch Engrg & Bus Solutions, Ltd.

84
Objective
• The aim of this paper is to recognize two categories of drivers like Novice and
Older drivers who are at most risk and represent a major portion of accident rate.

85
Introduction
• Driver behavior- Experience, Age, Gender, Older, Drowsy, Rational, Young.
• Novice driver- Less attention, Slow response, Less hazard perceiving ability.
• Older driver – Aging can lead to physical action, glaring sensitivity, hearing ability.
• Age may not define as Novice & Older driver.
• Majorly accidents are caused by high speeding, environment condition, poor road
infrastructure, violating traffic rules, rash driving etc.

86
Novice Driver Identification(NDI)
• Novice drivers are identified by behaviors like poor steering control, frequent
unnecessary braking, and delayed reactions when cornering.
• Response - Poor compared to Experienced drivers,
• Hazard Perception – Novice drivers are not attentive to all the objects on the
road, and they have less perceiving ability.
• Vision measures – Experienced drivers scan more broadly than novice drivers.
• Experienced drivers had average horizontal vision range 42 degrees.
• Novice drivers had an average horizontal vision range 15 degrees.

87
Algorithm for NDI

Throttle
Steering position is A sudden release of
angle > Yes same as Yes accelerator pedal which
Predefined observed in may be followed by
threshold the previous brake pedal application
cycle

88
Older driver identification(ODI)
• Older drivers are identified through slow pedal-switching times and slower physical
reactions.
• Perception reaction process - four tasks.
1. Perception – 6 ms
2. Identification
3. Decision
4. Volition
• PRT(Perception Reaction Time).
• PRT is directly proportion to Age.
• For identifying it take approximately 30 ms more than young driver.
• On worst case it take approximately 100 ms more than young drivers

89
Algorithm for ODI
Pedal release
is detected
Yes
Start the timer

Pedal press is
detected

Yes
Stop the timer

Check the time taken

90
Comparison of ND & OD
Novice driver(ND) Older driver(OD)
Driver does not provide the appropriate Driver does not provide the appropriate
response due to lack of experience. response due to age related inabilities
Driver does not know the right action to be Driver does know the right action to be
taken taken
Driver takes more time to complete Driver takes more time to complete
perception reaction process due to lack of perception reaction process due to slow
attention and less hazard perceiving ability physical movements and cognitive
measures
Driver misjudges the situation and so the Driver knows the right action, if he gets the
response provide is not appropriate and information. Less vision, hearing abilities
efficient. and cognitive measures affects receiving
Driver should be informed about the right information
action to be taken in each scenario. Driver should be assisted to receive the
information about the environment.

91
Adaptations of Braking Functions
• Adapting HBA(Hydraulic Brake Assist)
• Front Trajectory – It enables the driver to learn and become experienced.
• Adapting HBP(Hydraulic Brake Prefill)
• Adapting HHC(Hill Hold Control)
• Adapting Dashboard to Compensate Slower Saccade Latency and Glare
Sensitivity.

92
Paper- 4

Predict Vehicle Collision by TTC From Motion Using a

Single Video Camera.
Mehmet Kilicarslan et.al.

93
Objective
• The objective of this paper is the instantaneous computation of time-to-collision
(TTC) for potential collision only from the motion information captured with a
vehicle borne camera.

94
Introduction
• The method uses a single video camera mounted on the vehicle, which is simpler
and cheaper than using technologies like LiDAR or RADAR.
• The camera doesn’t rely on recognizing objects or their shapes; instead, it
focuses purely on the motion and size changes of objects.
• The TTC based on point tracking identify only motion in parallel to vehicle
direction.
• Other vehicles that approaching nonparallel to camera/vehicle heading direction
and vehicle on curved roads cannot be alarmed.
• Condensing the image to one dimensional data for fast computation.

95
Data
• Driving environment are full of lines, which can be categorized mainly three types
in video frames:
1. Horizontal lines on rear side of vehicle such as bumper, window, and top,
shadow and road marks on the ground.
2. Vertical lines on vehicles and background such as poles, and side objects.
3. Lines through depth on vehicle side view and adjacent lane marks.
• Applied algorithm onto naturalistic driving videos without accidents, and the
output shows the sensitivity of the algorithm to the moment that need braking.
• The videos have the resolution of 1280 x 720 pixels sampled at 30 frames per
second.
• The implementation has been done using Matlab2014b on windows 7.

96
Methodology

97
98
• Red arrow indicates motion direction of potential collision, green arrows mean safe
motion, and orange arrows mean centered motion direction requiring attention.
99
Classification Dangerous level

100
Results
• Various road types and driving actions, like sudden braking, lane-cutting, and
truck merging, were tested for collision detection.
• The system successfully alerted about potential collision dangers during testing
across different scenarios.
• This method is sensitive to front vehicle as far as 60m if scenes have a good
visibility.
• The environment changes may affect this method like, poor visibility at night,
heavy raining, sudden illumination changes, shadows from the roadside etc.
• In fast-moving traffic, matching and tracking points isn't practical for a simple
LiDAR map with 32 or 64 lines, as it's too limited for real-time driving.
• Also, the depth measurements from LiDAR can often give wrong time-to-collision
(TTC) values at the edges of vehicles.

101
Paper- 5

Impacts of advanced driver assistance systems on

commercial truck driver behaviour performance using
naturalistic data.
Jing Cao et.al

102
Objective
• This study investigated the impact of ADAS with respect to forward collision
warning(FCW), urban forward collision warning (UFCW), lane departure warning
(LDW), headway monitoring and warning (HMW), and speed limit indicator(SLI)
on commercial truck drivers’ behaviours using naturalistic data.
• By using one-way analysis of variance(ANOVA) method.
• At last, the purpose of this study was to examine the impact of ADAS on the
driving performance of commercial truck drivers based on naturalistic data.

103
Methodology

• 137 male commercial truck drivers from

four different freight companies.
• Age ranges from 22-55 years old.

104
Summary
• Under the condition of driving speed at 30-200 km/h, the FCW warns up to 2.7 s
in advance.
• If the driving speed is around 0-30 km/h, the UFCW warns up to 2 s in advance.
• LDW activates if the speed greater than 55 km/h.
• PCW generated when calculated time to collision with a pedestrian or cyclist is
less than 2 s.
• HMW monitors distance from the vehicle ahead and issues alerts when headway
is less than 1.4 s.
• SLI scans for speed limits signs along road and assists driver in keeping in that
limit.

105
Three stage test

106
Analysis- Results

107
Conclusion
• ADAS improved truck drivers' safety, especially with headway keeping and lane
departure compliance, but had limited impact on low-speed collision avoidance.
• Experienced truck drivers didn’t respond much to forward collision warnings at
low speeds.
• ADAS showed positive effects on overall safety, but more research is needed on
different types of commercial drivers.
• Improving truck driver training and supervision with ADAS can enhance traffic
safety and reduce energy consumption.

108
List of objectives
1. To analyze the impact of Advanced Driver Assistance Systems (ADAS) and Driver
Monitoring Systems (DMS) on improving driver safety across different
demographic groups and driving behaviors.
2. To evaluate the long-term behavioral changes in drivers induced by continuous
exposure to ADAS and DMS, focusing on the adaptation process and potential risk
compensation behaviors.
3. To determine the effectiveness of ADAS and DMS in reducing incidents of hard
braking, speeding, and distracted driving, particularly on different types of road
geometries (e.g., curves, intersections, highways).
4. To develop an integrated model that correlates driver behavior, road geometry, and
the effectiveness of warning alerts to identify critical safety combinations that need
further research and development.
5. To conduct a comparative study on the differential effects of ADAS and DMS on
novice versus experienced drivers, with a focus on the potential for these systems
to bridge the gap in driving skills.

109
• Random Forest machine learning algorithm- for multiple features, they can help
identify critical decision points in ADAS/DMS, such as when to issue an alert to a
driver.
• K- Nearest Neighbors (KNN) – This can be classifying when spatial or similarity-
based predictions are relevant. It will predict based on historical driving data.
• For yes/no questions- I would like to choose either Logistic Regression or
Decision Trees, that predict simple yes/no outcomes based on input features
such as speed, distance, driver alertness etc.

110