In smart cities, the frequency of traffic collisions has been rising daily.
More than half of all traffic-related deaths and injuries happen to people most vulnerable on the road. According to the World Health Organization report published
in 20182022, 1.3 million people died due to accidents on the road due to vehicles, and
30 to 50 million were injured. To address this issue, we present the road safety intelligent transport system (RSITS) based on deep federated learning-assisted fog cloud
networks. RSITS offers mobile LiDAR sensors and vehicle LiDAR sensors enabled
applications to alert road safety mechanisms. To deal with the complex features of
road safety, we trained the large pedestrian and vehicle detection dataset on different
road safety fog servers and aggregated them on the centralized cloud. To ensure that
constraints such as safety, the accuracy of alarms, response times, security, and deadlines are met, we present a deep federated learning scheduling scheme (DFLSS) that
consists of different components: Initially, we bound all applications so that emergency tasks, such as moving an object within 5 meters, should be processed locally
with the minimum response time. Due to resource constraints and the limitations of
devices, other tasks of applications are offloaded to the centralized cloud for processing. To ensure security, each computing node must encrypt and decrypt data before
offloading and processing it in DFLSS. Simulation results show that the proposed
DFLSS outperformed all existing approaches regarding accuracy, response time, and
deadline for road safety applications.
Breaking the Kubernetes Kill Chain: Host Path Mount
RSITS: Road safety Intelligent Transport System in Deep Federated Learning Assisted Fog Cloud Networks
1. Entitled: RSITS: Road safety Intelligent
Transport System in Deep Federated Learning
Assisted Fog Cloud Networks
Presented by Authors,
Senior Researcher, Dr. Abdullah Lakhan
School of Economics, Inovations and Technology
Kristiania University College, Oslo, Norway
Presented to
MobiWis 2023
The 19th International Conference on Mobile Web and Intelligent
Information Systems, Marrakech, Morocco
Monday, 14 August 2023
2. Outlines
• Introduction and Motivation of RSITS
• Research Applications
• Research Contributions
• Proposed Architecutre
• Methodology
• Results and Discussion
3. Introduction of Road Safety Intelligent
Transport System (RSITS)
RSITS Motivation
• Road safety is a critical concern in the
development of smart cities,
especially when multiple public
transports share the same pathways
with pedestrians.
• RSITS refers to the integration of
information and communication
technologies into transportation
systems to enhance safety, efficiency,
and sustainability at road side for End
Users.
• End users widely exploited the public
transports in Norway.
RSITS Scenario in OSLO
4. Applications of ITS for Road Safety
Applications of RSITS
• Collision Avoidance Systems
• Traffic Management and Control
• Connected Vehicles
• Pedestrian and Cyclist Safety
• Emergency Response
• Speed Management
• Road Safety and Vehicle Detection
5. Technologies and Methods for RSITS
• IoT Sensors
• Fog and Cloud Computing
• Wireless Communications
• Machine Learning->Neural Network, Federated Learning
• Security
• Dataset-> Vehicles, Pedestrian, Road side unit infrastructure
6. Research challenges
• Seasonal Challenges: In Norway, harsh weather conditions such as heavy snowfall
and low visibility during the winter can pose significant challenges for pedestrian
and vehicle detection systems. Snow-covered roads and reduced light can affect
the accuracy of sensors like cameras and LiDAR, which are commonly used for
detection.
• Pedestrian Behavior: Pedestrian behavior can be unpredictable, making it
challenging for detection systems to anticipate their movements. Systems need to
be designed to handle sudden movements, jaywalking, and interactions with
vehicles.
• Data Privacy: As with any intelligent transport system, the issue of data privacy is
crucial. Systems that involve tracking pedestrians and vehicles must adhere to
strict privacy regulations and ensure that data is anonymized and secured.
7. Research Solution
• To address all questions mentioned above, this study presents secure
pedestrian and vehicle collision avoidance road safety intelligent
transport system (RSITS) in deep reinforcement learning assisted fog
cloud networks. We present secure pedestrian and vehicle collision
avoidance by Road Safety Intelligent Transport Systems (RSITS) in
deep reinforcement learning assisted fog networks.
8. Contributions of RSITS
• We present mobile LiDAR sensors and vehicle LiDAR sensor-enabled applications to alert
road safety mechanisms. To deal with the complex features of road safety, we trained the
large pedestrian and vehicle detection dataset on different road safety fog servers and
aggregated them on the centralized cloud.
• We have collected different dataset images of different public transport modes, vehicles,
pedestrians, and routes, about 500000 images. We named this dataset the Road Safety
Dataset (RS-Dataset). We divided the road safety dataset into sub-datasets and trained
and validated them on different servers. These sub-datasets have pre-trained and
realtime images in the proposed RSITS.
• We are considering different constraints such as security, deadline, response time, and
an accurate safety alarm. So, we developed the deep federated learning scheduling
scheme (DFLSS) to execute all applications based on their constraints.
• We present a research product-based simulation tool with source code and a dataset for
researchers to further analyze and develop road safety in smart cities.
18. Conclusion
Applications and Dataset
The study aimed to lessen accident risks by identifying vehicles' intentions, distances, and
speeds, leading to swift safety alerts. They gathered a diverse dataset comprising
approximately 500,000 images of various public transport vehicles, pedestrians, and
routes. We designed the mobile and vehicle enabled road safety applications for end users
and achieved more good results as compared to existing applications.
Methods
This dataset was segmented and used for training and validation on separate servers.
These subsets contained both pre-trained and real-time images within the proposed RSITS.
To address factors like security, deadlines, response times, and accurate object prediction,
a deep federated learning scheduling scheme (DFLSS) was devised. Simulation results
indicated that the DFLSS surpassed existing methods with a 5% accuracy improvement,
15% faster response time, and a 20% better deadline achievement for road safety
applications.
19. Laboratory
• We have designed applied research based on ubiquitous Lab:
https://ubiquitouslab.no/
• Partner Companies: Reuter, Kogenta, and transport data company
• Team Members: Professor Tor Morten Groenli, Abdullah Lakhan,
Avnish Jat.
• Abdullah.Lakhan@kristiania.no
• +4793018086
20. Thanks
• Thank you so much for your valuable time. Any question welcome.