TITLE:

An Innovative Anomaly Driving Detection Strategy for Adaptive FCW of

CNN Approach

Raja Mariatul Qibtiah Binti Raja Jaapar

Industrial Automation Dept.
Universiti Kuala Lumpur, MFI
Selangor, Malaysia
UNIVERSITI KUALA LUMPUR - MALAYSIA FRANCE
INSTITUTE
 INTRODUCTION
FORWARD COLLISION WARNING IS AN EMERGING SAFETY TECHNOLOGY
Uses cameras, radar, laser, and sensor to monitor the road ahead Each driver has specific characteristics which drastically affect his
and alert the driver decision and reactions in different driving situations

THE DISTRACTION DETECTION MODEL IS MEASURED WITH SEVERAL PARAMETERS.

Gas pedal position, brake flag, yaw rate, turn signal, longitudinal Probability of false warning detection rate remains high due to
acceleration, range rate, speed, heading, and range insufficient input to support existing parameters used

BUILT-IN SYSTEM'S FLEXIBILITY IS MOST SIGNIFICANT FOR BETTER OUTPUT IN AN

ADAPTIVE BEHAVIORS TO DETECT THE WARNING
Self-learning algorithms for driver characteristics based on Classical method that use mathematical models to represent the
recursive least square method (LSM) relationship between the state variables of the host vehicle.
 INTRODUCTION
INADEQUATE INFORMATION ON HOW TO DETERMINE THE APPROPRIATE TIME FOR A
WARNING IN on
Focused ORDER
using TO GET HIGH
machine EFFECTIVENESS AND ACCEPTANCE BY THE DRIVER.
learning
Using an advanced adaptive FCW algorithm as
identification to reduce false alarm of collision
inputs.
warning system

IMPLEMENTING CLASSIFIER INTO FORWARD-COLLISION WARNING SYSTEM (FCW)

FRAMEWORK

Non-Adaptive Model Adaptive Model

 AN ADAPTIVE FORWARD VEHICLE COLLISION MODEL

• Convolutional Neural Network:

 The initial concept of neural network was developed by scientists from the human nervous system.
 Used to improve the driving efficiency as of this model lead to the improvement on artificial
intelligence.
• Support Vector Machine:
 From the enhancement of theoretical approach, SVM has been developed such that it can be used in
multiple diagnostics to build more complex models.
 SVM performs well in most domains and tasks such that they are unaffected to dimensional curses and
can efficiently handle large-scale problems in samples and variables.
• Multi-layer Perceptron Model:
 Used in facial recognition to cover the human error in the aspect of classification.
 Feed-forward artificial neural network model that pulls a set of input data into an appropriate set of
outputs.
 D. model et al. J. Chen et al., Y. Sugano, F. Dong et al.
Alnajar, Y. Zhang, Farah Nadia - A comprehensive study of
- Concerned with eyes
tracking system which - Writes that the same driver monitoring systems can
detected eyes feature concept without calibration be found in this research
extraction. and requires more input while - Consequently, it is important
adjusting certain parameters. for the Advanced Driver
- Found a common
method by extending - The applied correction Assistance System (ADAS)
the detection range to process to achieve good - To invasive these distractions
a remote eye-tracking accuracy and performance to warn the driver or guided if
system a dangerous situation
happened.
 Ohno et al. Boer et al. Daimler Chrysler, W.Bouslimi et al
- Established a driver -Adopted some artificial - Who used the naive Bayesian
model using neural intelligence methods to network classifier to model the
networks to predict establish driver models from driver behavior
the driving behavior the perspective of human - The rear-end collision system
of the driver when psychological and needs to provide the driver with
using the adaptive physiological mechanisms warning information. At the same
cruise control system -Driver model that uses a time, the system's automatic
- The output of the three-layer structure to control of the vehicle will also
model is the speed describe driver behavior based affect the driver's driving
and steering angle. on knowledge, rules, and skills behavior
 Japan’s Mazda Genya Abe et al. The University of Michigan
Corporation - Impact of different alarm in the United States
- Uses driving simulator timings of the system on the - Studied the similarities
experiments and data driver’s braking response and and differences between
statistics methods to the driver’s trust in the system. manual driving with a
study the impact of the - Different alarm experiments driving assistance system,
frequency, quality and were designed mainly in the - Provides a research basis
heuristics of rear-end form of driving simulator for the development of
collision alarms on the experiments and the driver’s driving assistance systems
driver’s reaction time response was obtained through and the impact of the
and response actions statistical data and analysis system on traffic conditions
 NECESSITY AND APPLICATION OF DRIVING DETECTION CLASSIFIERS

Factors found to be influencing fatigue have been explored in

several studies particularly on drivers who were involved in long-
distance passenger.
Driving distraction detection are divided into two factors:
 Distraction detection from physiological attributes.
 Distraction detection from visual face features.
Characteristics of the occurrence for distraction to driving.
 Head position to identify tiredness.
 Eye condition to decide drowsiness.
 Degree of opening and closing of the mouth
• SVM classifier :
Able to indicate fatigue according to the actual eyes of the driver and size of mouth which
produced high accuracy due to characteristics of the inputs taken.

• Random Forest Classifier :

Developing monitoring system related with identifying the condition of vehicular drivers to
detect fatigue.

• Deep Cascade Convolutional Neural Network Classifier :

Identifying the visual face state from video and detect the eyes through Dlib toolkit
application.

• Multi-Task Cascaded Convolutional Neural Network Classifier :

Developing for facial detection. The content with three sub-network which can be more
effective in terms of algorithm’s speed due to less number of filters to run the system.
 SIGNIFICANCE OF RESEARCH
• This study's findings will redound to the benefit of society.
• Considering that the development of enhancing driver assistance techniques
in forwarding forward collision warning system technology.
• Minimize our nation’s dependency on foreign technology in the vision
engineering field.
• The outcome can be applied to detect anomalies or distractions of the drivers
in order to:
- Increase road safety levels.
- Reduce financial implications.
• This is in line with the efforts of the Malaysian government to drive the
country towards knowledge-based society and economy.
 CONCLUSION
- Minimize the rate of false and drivers receive the fastest early warning.
-Provides insights into the efficiency of the CNN system to extract driver eyes closeness,
considering its exploratory nature
-Assess whether the reaction is risky (drowsy or tired).

-Using the appropriate model to achieve the desired and reliable alert so that when the alarm is
automatically triggered, the driver does not feel disturbed
-Established that accidents that have happened must be stressed and concerns that are easier to
look at first before accidents occur

- Stage of eye closeness improves the driver's visibility and then sends it to an algorithm to
analyze and decide if the alarm should be switched on or off while an object can be identified in
the front.
- Underline the significant outcomes of using two input evaluation approaches, providing more
reliable and quality results, and being consistent with the use of high-tech computer vision
THANK YOU
AND
STAY SAFE