Ahamed / Thirunavukarasu / Nagarajan Recent Advancements in Computational Intelligence: Concepts, Methodologies and Applications (Part 2)

Concepts, Methodologies and Applications (Part 2)
1. Auflage 2026
ISBN: 979-8-89881-288-1
Verlag: De Gruyter
Format: EPUB
Kopierschutz: 0 - No protection

E-Book, Englisch, 280 Seiten

Recent Advancements in Computational Intelligence: Concepts, Methodologies and Applications (Part 2) explores cutting-edge developments in computational intelligence (CI), a dynamic field at the intersection of artificial intelligence, machine learning, and data science. Emphasising biologically inspired techniques such as neural networks, fuzzy logic, swarm intelligence, and evolutionary algorithms, the book demonstrates how CI solves real-world challenges involving uncertainty, ambiguity, and incomplete information. Aligned with Industry 4.0, it bridges theoretical foundations with practical applications across diverse domains, including healthcare, autonomous systems, cybersecurity, and smart computing.
Structured into thematic sections, the book covers Social Computing, High-Performance Computing, Network Science, Smart Computing, and Intelligent Communications. Key chapters explore deep learning for autonomous vehicles, AI-driven healthcare diagnostics, graph theory for cybersecurity, edge computing in IoT, and reinforcement learning for generative AI. Case studies highlight innovations such as myelin segmentation in pediatric MRI, personalised marketing AI, and secure password generation with multi-factor authentication. Contributions from global experts offer a balanced blend of theoretical rigour and applied research.
Key Features
Comprehensive coverage integrating CI concepts with emerging trends like generative AI and edge computing
Real-world case studies demonstrating CI's transformative potential in industry
An interdisciplinary approach connecting computer science, engineering, and healthcare
Practical methodologies for researchers and implementable strategies for practitioners
Target Readership
Researchers, academics, students and professionals in AI, computer science, and data engineering

Autoren/Hrsg.

Ahamed, Bazeer

Thirunavukarasu, Ramkumar

Nagarajan, Rajganesh

Fachgebiete

• Mathematik | Informatik EDV | Informatik EDV & Informatik Allgemein

Leseproben

Deep Neural Networks for Autonomous Vehicles: A Review of Advances, Challenges and Future Directions

Arul Prasath A.1, *, K. Navaz1

1 Department of Computer Science & Engineering, Vel Tech Rangarajan Dr.Sagunthala R&D Institute of Science and Technology, Avadi, Chennai, India

Abstract

Deep Neural Networks (DNNs) have become a vital technology in the quick development of autonomous cars, enabling vision, decision-making, and control systems. This has created new avenues for exploring intelligent traffic safety, comfortable travel, and smart roadways. Because of its enormous potential to decrease traffic accidents and injury to people, autonomous vehicles have recently become a very attractive study topic. However, object detection, scene identification, lane recognition, and additional challenges are crucial for self-driving cars to solve. The development, difficulties, and probable future directions of DNNs for driverless vehicles are thoroughly examined. Perception tasks, sensor fusion, safety issues, decision-making algorithms, and the incorporation of DNNs into practical autonomous driving systems are just a few of the many areas it addresses. This review attempts to offer insights and suggestions for more breakthroughs in the field through the investigation of the most recent state-of-the-art research and highlighting significant difficulties.

Keywords: Autonomous driving and vehicles, Deep neural networks, Decision-making, Object detection, Sensor fusion.

* Corresponding author Arul Prasath A.: Department of Computer Science & Engineering, Vel Tech Rangarajan Dr.Sagunthala R&D Institute of Science and Technology, Avadi, Chennai, India; E-mail: [e-mail]

INTRODUCTION

A recent technical analysis claims that the National Highway Traffic Safety Administration (NHTSA), nearly 95% of traffic collisions are prone to human-made errors [1]. Autonomous vehicles play a major role in creating a suitable driving scheme that avoids collisions, reduces emissions, aids in transferring the mobility-impaired, and decreases stress related to driving [2]. ADSs are expected to provide better yearly social benefits of almost $800 billion by 2050, if extensive implementation is successful, it will help in reducing traffic congestion,

the number of traffic fatalities, and energy consumption, and increasing productivity as a result of smarter driving time. Autonomous driving technologies work to reduce these fatalities while also enhancing traffic flow and overall driving safety. Therefore, reducing the number of fatalities caused by human drivers can be avoided by building an autonomous driving system that would be precise, and reliable to function in a variety of environmental situations, thereby providing quick and efficient decision-making schemes for speed driving at the same time adhering to necessary safety requirements [3].

Autonomous vehicles (AVs) should comprehend driving environments, which include other participating vehicles, bicycles, pedestrians, lane identity information, and constraints like the speed-limit zone given on streets, in order to make them more reliable. The driving environment along with its associated locations is better understood with the aid of perception technologies like object detection. The perception system must provide exact and reliable information on the road conditions and be durable enough to function in inclement weather, such as rain, snow, haze, and driving at night, and be equipped with real-time decision-making capabilities to accommodate high-speed driving. Additional advantages of autonomous vehicles have been noted, including higher fuel economy, decreased pollution, automobile sharing, increased efficiency, and more efficient traffic management [4-6].

The University of Bundeswehr Munich and Carnegie Mellon University revealed several early autonomous vehicle studies on driving on highways and structured areas respectively in the 1980s [7, 8]. Since then, research into AVs has continued to advance due to initiatives like DARPA Grand Challenges [9, 10]. Aside from academic institutions, automakers and IT firms have also conducted research to create their AVs. Due to this, modern automobiles now include a variety of advanced driver assistance systems that give them some autonomy, including Collision Warning, Lane Assistance, and adaptive cruise control (ACC)technologies. These developments not only make driving safer and easier in contemporary cars, but they also open the path for fully functional autonomous cars that do not need drivers to operate them.

Numerous discoveries and useful applications in numerous fields have been made possible with the research advances related to machine learning and deep learning techniques [11-15]. The automobile industry and the creation of completely autonomous cars are the areas where ML has the biggest impact. Many autonomous vehicle subsystems, including localization and mapping, sensor fusion, perception, and path planning, employ ML techniques [16-19]. The present trend is the evolution of multiple automobile platforms concurrent with work in line with fully autonomous commercial vehicles. The sensors-based perception subsystem, path planning subsystem, and vehicle control schemes are the four main building parts which have been shown in Fig. (1), depicting the overall sketch of the autonomous driving system. Deep learning has various advantages for controlling vehicles. Deep learning is particularly suited to mitigate issues in varied and intricate environments because of its capacity to self-optimize its behaviour from information and respond to that dynamically changing driving scenarios [20-22]. Deep learning algorithms help in predicting the intended behaviour and train the ADS platform to perform effectively and generalise to new contexts through learning [23, 24]. This eradicates the need for modifying each parameter repeatedly while attempting to preserve performance in all possible circumstances. These factors have contributed to the recent surge in demand for deep learning for AV control.

Fig. (1))
Generic block diagram of AV system

Several various sensors put on the vehicle are used to sense the environment. A perception block converts sensory data into meaningful information after processing the input from the sensors. The perception block’s output is used by the planning subsystem for both long- and short-range path planning as well as behaviour planning. The autonomous car receives commands from the control module and makes sure it travels along the course laid out by the planning subsystem. To perform a proper steering control scheme for ADS, an end-to-end DNN is developed for autonomous driving and employs images from the camera as an input and steering angle forecasts as an output as shown in Fig. (2). End-to- end learning describes the training of artificial networks from the start point to the end point without any support from humans. End-to-end learning's goal is to automatically teach the system internal models of the required processing stages, including identifying relevant road features, entirely based on the signal being provided.

Fig. (2))
End-to-End AV System

Although a wide range of other approaches have been utilised to address AV control via DL techniques, there is currently a paucity of analysis and comparison of these various methods. By examining and evaluating the deep learning techniques for vehicle control, this review tries to close this gap in the literature. Additionally, this survey will assess the existing state of the subject, pinpoint the key research difficulties, and offer suggestions for the course of future investigation. Each domain was examined by looking at various approaches and techniques and comparing and contrasting their benefits, drawbacks, results, and significance. The remainder of this survey includes six sections. Section II covers the technique for understanding the environment. Section III discusses the sensor fusion techniques of AVs based on DNNs. Section IV investigates the adversarial attacks and other safety and reliability issues for Avs. Section V describes decision-making control techniques related to Avs. Section VI presents the research challenges and future perspectives with respect to autonomous driving systems (ADS). Section VII concludes with final remarks on the development of AV systems.

To sum up, this work identifies the potential challenges in addressing end-to-end ADS right from collecting data using sensors like cameras, LiDAR, and radar that captures the surroundings of the vehicle. Next, the perception module comes into play. It identifies objects, lane markings, and obstacles around. Localization is crucial too, it pinpoints the vehicle's exact spot using GPS mixed with other sensor data (sensor fusion). After that, path planning happens. This part generates a safe route, considering changing...