Computer Science and Mathematics

Sort by

Article
Computer Science and Mathematics
Computer Vision and Graphics

Pooja Kumari,

Björn Van Marwick,

Johann Kern,

Matthias Rädle

Abstract:

Three-dimensional (3D) cellular models, such as spheroids, serve as pivotal systems for understanding complex biological phenomena in histology, oncology, and tissue engineering. In response to the growing need for advanced imaging capabilities, we present a novel Multi-Modal Raman Light Sheet Microscope designed to capture elastic (Rayleigh) and inelastic (Raman) scattering, along with fluorescence signals, in a single platform. By leveraging a shorter excitation wavelength (532nm) to boost Raman scattering efficiency and incorporating robust fluorescence suppression, the system achieves label-free, high-resolution tomographic imaging without the drawbacks commonly associated with near-infrared modalities. An accompanying Deep Image Prior (DIP) seamlessly integrates with the microscope to provide unsupervised denoising and resolution enhancement, preserving critical molecular details and minimizing extraneous artifacts. Altogether, this synergy of optical and computational strategies underscores the potential for in-depth, 3D imaging of biomolecular and structural features in complex specimens and sets the stage for future advancements in biomedical research, diagnostics, and therapeutics.

Article
Computer Science and Mathematics
Algebra and Number Theory

Ferdi Ferdi,

Amir Kamal Amir,

Andi Muhammad Anwar

Abstract: Kac-Moody Lie algebra is a Lie algebra associated with Cartan matrix generalized over real or complex field. This research aims to define Kac-Moody Lie algebra in quaternion by using the concept of Quaternification of Lie algebra. From some previous research, the definition of Kac-Moody Lie algebra over real or complex field is divided into two, namely the definition of standard and reduced Kac-Moody Lie algebra. To obtain both definitions, one additional definition is needed, namely the universal Kac-Moody Lie algebra. So, to define Kac-Moody quaternion Lie algebra, three constructions are needed, namely the construction of universal Kac-Moody quaternion Lie algebra, standard Kac-Moody quaternion Lie algebra, and reduced Kac-Moody quaternion Lie algebra. The results of this paper obtained the definition of universal Kac-Moody quaternion Lie algebra, standard Kac-Moody quaternion Lie algebra, and reduced Kac-Moody quaternion Lie algebra.
Article
Computer Science and Mathematics
Robotics

Yinlong Liu

Abstract: Rotation motion in a three-dimensional physical world refers to an angular displacement of an object around a specific axis in $\mathbb{R}^3$. It is typically formulated as a non-linear and non-convex motion due to the nonlinearity and nonconvexity of $\mathbb{SO}(3)$. However, this paper proposes a new perspective that the 3D rotation motion can be expressed by a linear system without dropping any constraints and increasing any singularities. Moreover, two frequent cases, i.e., $\angle\left(\mathbf{R}\boldsymbol{x},\boldsymbol{y}\right)=0$ and $\angle\left(\mathbf{R}\boldsymbol{x},\boldsymbol{y}\right)=\frac{\pi}{2}$, in computer vision and robotics that can be expressed linearly are deeply discussed in this paper.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Georgios Spanos,

Antonios Lalas,

Konstantinos Votis,

Dimitrios Tzovaras

Abstract: Cooperative, Connected, and Automated Mobility (CCAM) is set to play a key role in the future of transportation, contributing to the achievement of sustainable development goals. Moreover, Artificial Intelligence (AI), a transformative technology with applications across various industries, can significantly enhance CCAM operations. Additionally, passenger demand forecasting, a critical aspect of mobility research, will become even more essential as CCAM adoption continues to grow in the next years. Therefore, the present research study in order to deal with the issue of passenger demand forecasting in CCAM, proposes the Principal Component Random Forest (PCRF) methodology, which is based on AI as it leverages a well-established statistical methodology such as the Principal Components Analysis with a flagship traditional machine learning technique, which is Random Forest. The application of PCRF in four European pilot sites within the EU-funded SHOW project demonstrated its high accuracy and effectiveness.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Aziida Nanyonga,

Hassan Wasswa,

Keith Joiner,

Ugur Turhan,

Graham Wild

Abstract:

The timely identification of probable causes in aviation incidents is crucial for averting future tragedies and safeguarding passengers. Typically, investigators rely on flight data recorders, yet delays in data retrieval or damage to the devices can impede progress. In such instances, experts resort to supplementary sources like eyewitness tes- timonies and radar data to construct analytical narratives. Delays in this process have tangible consequences, as evidenced by the Boeing 737 MAX accidents involving Lion Air and Ethiopian Airlines, where the same design flaw resulted in catastrophic outcomes. To streamline investigations, scholars advocate for natural language processing (NLP) and topic modeling methodologies, which organize pertinent aviation terms for rapid analysis. However, existing techniques lack a direct mechanism for deducing probable causes. Bridging this gap, this study proposed a transformer-based model for predict- ing likely causes from raw text narrative inputs, leveraging advancements in long-input transformers. By training the model on comprehensive aviation incident investigation reports like those from the National Transportation Safety Board (NTSB), the proposed approach exhibits promising performance across key evaluation metrics, including Bilin- gual Evaluation Understudy (BLEU) with (M=0.727, SD=0.33), Latent Semantic Analysis (LSA similarity) with (M=0.696, SD=0.152), and Recall Oriented Understudy for Gisting Evaluation (ROUGE) with a precision, recall and F-measure scores of (M=0.666, SD=0.217), (M=0.610, SD=0.211), (M=0.618, SD=0.192) for rouge-1, (M=0.488, SD=0.264), (M=0.448, SD=0.257), M=0.452, SD=0.248) for rouge-2 and (M=0.602, SD=0.241), (M=0.553, SD=0.235), (M=0.5560, SD=0.220) for rouge-L, respectively. This demonstrates its potential to expedite investigations by promptly identifying probable causes from analysis narratives, thus bolstering aviation safety protocols

Article
Computer Science and Mathematics
Computer Vision and Graphics

Seok-Woo Jang,

Limin Yan,

Gye-Young Kim

Abstract: In this study, we propose a dynamic scene deblurring approach using a deep supervised attention network. While existing deep learning-based deblurring methods have significantly outperformed traditional techniques, several challenges remain: 1) Invariant weights: Small conventional neural network (CNN) models struggle to address the spatially variant nature of dynamic scene deblurring, making it difficult to capture the necessary information. A more effective architecture is needed to better extract valuable features; 2) Limitations of standard datasets: Current datasets often suffer from low data volume, unclear ground truth (GT) images, and a single blur scale, which hinders performance. To address these challenges, we propose a multi-scale, end-to-end recurrent network that utilizes supervised attention to recover sharp images. The supervised attention mechanism focuses the model on features most relevant to ambiguous information as data are passed between networks at difference scales. Additionally, we introduce new loss functions to overcome the limitations of the peak signal-to-noise ratio (PSNR) estimation metric. By incorporating a fast Fourier transform (FFT), our method maps features into frequency space, aiding in the recovery of lost high-frequency details. Experimental results demonstrate that our model outperforms previous methods in both quantitative and qualitative evaluations, producing higher-quality deblurring results.
Article
Computer Science and Mathematics
Computer Science

Godwin Olaoye

Abstract: The Unified Extensible Firmware Interface (UEFI) is an essential component of modern computing systems, providing a flexible and efficient interface between the operating system and firmware. One key feature of UEFI is Capsule Updates, which allow the secure delivery and installation of firmware updates, ensuring that the system can be kept up to date with the latest fixes and improvements. However, this update mechanism can also be vulnerable to malicious attacks, particularly Man-in-the-Middle (MITM) attacks and other threats that target the integrity of the firmware update process. Secure Boot, a foundational security feature in UEFI, is designed to counter these vulnerabilities by ensuring that only trusted firmware is allowed to execute during the system’s boot process. Secure Boot verifies the authenticity of the firmware before it is loaded, checking for valid digital signatures to confirm that the firmware has not been tampered with. This prevents the installation of unauthorized or malicious updates that could compromise the system’s security by preventing attackers from introducing harmful code during the UEFI Capsule Update process.In addition to protecting against MITM attacks, Secure Boot helps guard against other risks, such as rootkits, malware, and unauthorized firmware downgrades. It works alongside other security measures, including digital signatures for update files, cryptographic hashing to verify the integrity of firmware, and Trusted Platform Module (TPM) to securely store cryptographic keys. Together, these technologies create a robust defense system that ensures firmware updates are authentic, intact, and safe from exploitation.This paper delves into the critical role of Secure Boot in protecting UEFI Capsule Updates, focusing on its ability to enforce trusted firmware environments, preventing malicious actors from gaining control over system firmware. Furthermore, the paper explores how Secure Boot integrates with other hardware-based security features, forming a multi-layered approach to safeguarding the entire update lifecycle. By leveraging these features, organizations can significantly reduce the risk of firmware-based attacks and ensure that the UEFI Capsule Update process remains secure and reliable.Ultimately, Secure Boot plays a vital role in fortifying the firmware update process against evolving cyber threats. As UEFI Capsule Updates are an essential part of modern systems' maintenance, implementing Secure Boot ensures that updates are both authentic and secure, protecting the system from potential vulnerabilities and providing a foundation for trustworthy and resilient computing.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Mona Maze,

Samar Attaher,

Mohamed O. Taqi,

Rania Elsawy,

Manal M.H. Gad El-Moula,

Fadl A. Hashem,

Ahmed S. Moussa

Abstract:

This study investigated Land Use and Land Cover (LULC) classification east of the Nile Delta, Egypt, using Sentinel-2 bands, spectral indices, and Sentinel-1 data. The aim was to enhance agricultural planning and decision-making by providing timely and accurate information, addressing limitations of manual data collection. Several Machine Learning (ML) and Deep Learning (DL) models were trained and tested using distinct temporal datasets to ensure model independence. Ground truth annotations, validated against a reference Google satellite map, supported training and evaluation. XGBoost achieved the highest overall accuracy (94.4%), surpassing the Support Vector Classifier (84.3%), while Random Forest produced the most accurate map with independent data. Combining Sentinel-1 and Sentinel-2 data improved accuracy by approximately 10%. Strong performance was observed across Recall, Precision, and F1-Score metrics, particularly for urban and aquaculture classes. Uniform Manifold Approximation and Projection (UMAP) technique effectively visualized data distribution, though complete class separation was not achieved. Despite their small size, road area predictions were reliable. This research highlights the potential of integrating multi-sensor data with advanced algorithms for improved LULC classification and emphasizes the need for enhanced ground truth data in future studies.

Concept Paper
Computer Science and Mathematics
Mathematics

Bharath Krishnan

Abstract: Let n ∈ N and suppose function f : A ⊆ R^n → R, where A and f are Borel. We want a satisfying average for all pathological f (e.g., everywhere surjective f whose graph has zero Hausdorff measure in its dimension) taking finite values only. If this is impossible, we wish to average a nowhere continuous f defined on the rationals. The problem is that the expected value of these examples of f , w.r.t the Hausdorff measure in its dimension, is undefined. We fix this by taking the expected value of chosen sequences of bounded functions converging to f with the same satisfying and finite expected value. Note, “satisfying” is explained in the leading question which uses rigorous versions of phrases in the former paragraph and the “measure” of a bounded functions’ graph which involves minimal pair-wise disjoint covers of the graph with equal ε measure, sample points from each cover, paths of line segments between sample points, the lengths of the line segments in the path, removed lengths which are outliers, remaining lengths which are converted into a probability distribution, and the entropy of the distribution. We also explain “satisfying” by defining the actual rate expansion of a bounded functions’ graph and also “the rate of divergence” of a bounded functions’ graph compared to that of other bounded functions’ graphs.
Article
Computer Science and Mathematics
Signal Processing

Sayantan Bhattacharya,

Dimitris M. Christodoulou,

Silas G. T. Laycock

Abstract: The broad point spread function of the NuSTAR telescope makes resolving astronomical X-ray sources a challenging task, especially for off-axis observations. This limitation has affected the observations of the high-mass X-ray binary pulsars SXP 15.3 and SXP 305, in which pulsations are detected from nearly overlapping regions without spatially resolving these X-ray sources. To address this issue, we introduce a deconvolution algorithm designed to enhance NuSTAR’s spatial resolution for closely-spaced X-ray sources. We apply this technique to archival data and simulations of synthetic point sources placed at varying separations and locations, testing thus the algorithm’s efficacy in source detection and differentiation. Our study confirms that on some occasions when SXP 305 is brighter, SXP 15.3 is also resolved, suggesting that some prior non-detections may have resulted from imaging limitations. This deconvolution technique represents a proof of concept test for analyzing crowded fields in the sky with closely-spaced X-ray sources in future NuSTAR observations.
Article
Computer Science and Mathematics
Computational Mathematics

Sheed Iseal

Abstract: UEFI (Unified Extensible Firmware Interface) Capsule Update Mechanisms play a pivotal role in modern system firmware updates by providing a standardized method for applying firmware patches and updates. These mechanisms enable the installation of new firmware versions on devices that rely on UEFI, improving system functionality, security, and stability. A UEFI Capsule is a structured binary package containing update data and metadata, which is used to modify the firmware environment of a system. The Capsule Update process involves securely verifying the update's integrity and authenticity, ensuring that only trusted, non-tampered updates are installed.This abstract outlines the key components of UEFI Capsule updates, including the update format, the encapsulation of firmware data, and the critical role of security measures like authentication and integrity checks. Through a detailed review of the process, we examine how UEFI Capsule updates contribute to streamlined firmware management while mitigating security risks such as firmware compromise or unauthorized updates. Additionally, this paper highlights the integration of advanced technologies, such as digital signatures and secure boot mechanisms, to ensure a trusted and reliable update procedure. The discussion further emphasizes the importance of secure update practices in addressing the challenges faced in maintaining the security and reliability of modern computing platforms.By understanding UEFI Capsule Update Mechanisms, this work aims to provide insights into their essential role in the evolution of firmware management, addressing both technical challenges and security concerns associated with the update process.
Article
Computer Science and Mathematics
Computer Science

Godwin Olaoye

Abstract:

Unified Extensible Firmware Interface (UEFI) is a critical component in the boot process of modern computing systems, responsible for initializing hardware and loading the operating system. UEFI firmware updates, typically delivered in the form of "capsules," are essential for ensuring that a system operates with the latest security patches, bug fixes, and feature enhancements. However, as these updates are often transmitted over untrusted communication channels, they are vulnerable to Man-in-the-Middle (MitM) attacks. In such attacks, a malicious actor can intercept, modify, or inject harmful code into the firmware update, potentially compromising system security and integrity.This paper investigates the potential security risks associated with MitM attacks on UEFI capsule updates and proposes a set of robust countermeasures. We begin by analyzing the typical attack vectors, focusing on the vulnerabilities inherent in the process of transmitting update data between the system and update servers. Drawing from existing cryptographic protocols, we propose the use of strong encryption, digital signatures, and public-key infrastructures (PKIs) to ensure the authenticity, confidentiality, and integrity of UEFI capsule updates. These measures help verify that updates have not been tampered with during transit, preventing unauthorized code injection by attackers.Furthermore, we explore the role of Secure Boot and Trusted Platform Module (TPM) technologies in providing an additional layer of defense. By using secure boot processes, systems can verify that only authorized firmware updates are executed, thus preventing malicious code from being activated. Similarly, TPM can be employed to securely store cryptographic keys and validation certificates, reducing the risk of key compromise during the update process.We also delve into the concept of certificate pinning, where specific trusted certificate authorities are hardcoded into the UEFI firmware, making it more difficult for attackers to present fraudulent certificates during the update process. This approach significantly reduces the likelihood of MitM attacks exploiting misconfigured or compromised certificate chains.

Essay
Computer Science and Mathematics
Mathematics

Haoyuan Wang

Abstract: It is apologize to upload this incomplete draft since the time limitation, yet in this paper, we are going to give a new proof of the Eckart-Young-Mirsky Theorem which is crucial in machine learning, image and date processing etc.
Article
Computer Science and Mathematics
Computer Networks and Communications

Hojoon Lee,

Minjun Kim

Abstract: Efficient resource management in cloud-based web platforms is critical to maintaining performance and cost efficiency under dynamic and unpredictable workloads. This paper proposes a novel resource management framework that integrates predictive workload modeling, multi-tier autoscaling, and cost-aware optimization. The framework utilizes machine learning models to forecast workload patterns and coordinates resource allocation across application, caching, and storage tiers, ensuring minimal latency and optimal resource utilization. Experimental results demonstrate a 45% reduction in mean latency and a 30% decrease in total resource costs compared to traditional threshold-based autoscaling. The framework also improves resource utilization to 85% on average while halving the frequency of scaling actions, reducing operational instability. These outcomes highlight the effectiveness of the proposed approach in balancing performance and cost objectives in complex cloud environments. The proposed framework advances the state of the art in cloud resource management by addressing inter-tier dependencies and leveraging predictive analytics for proactive scaling. Its adaptability to diverse workload patterns and potential applicability to multi-cloud and edge computing scenarios make it a scalable and robust solution for modern web platforms.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Ashkan Farhadi

Abstract: Almost 70 years ago, Alan Turing predicted that within half a century, computers would possess processing capabilities sufficient to fool interrogators into believing they were communicating with a human. While his prediction materialized slightly later than anticipated, he also foresaw a critical limitation: machines might never become the subject of their own thoughts, suggesting that computers may never achieve self-awareness. Recent advancements in AI, however, have reignited interest in the concept of consciousness, particularly in discussions about the potential existential risks posed by AI. At the heart of this debate lies the question of whether computers can achieve consciousness or develop a sense of agency—and the profound implications if they do.Whether computers can currently be considered conscious or aware, even to a limited extent, depends largely on the framework used to define awareness and consciousness. For instance, IIT equates consciousness with the capacity for information processing, while the Higher-Order Thought (HOT) theory integrates elements of self-awareness and intentionality into its definition.This manuscript reviews and critically compares major theories of consciousness, with a particular emphasis on awareness, attention, and the sense of self. By delineating the distinctions between artificial and natural intelligence, it explores whether advancements in AI technologies—such as machine learning and neural networks—could enable AI to achieve some degree of consciousness or develop a sense of agency.
Article
Computer Science and Mathematics
Computer Vision and Graphics

Rafika Harrabi,

Slim Ben Chaabane,

Hassene Seddik

Abstract:

The recent advancements in technologies such as artificial intelligence (AI), computer vision (CV), and Internet of Things (IoT) have significantly extended various fields, particularly in surveillance systems. These innovations enable real-time facial recognition processing, enhancing security and ensuring safety. However, mobile robots are commonly employed in surveillance systems to handle risky tasks that are beyond human capability. In this context, we present a prototype of a cost-effective mobile surveillance robot built on the Raspberry PI 4, designed for integration into various industrial environments. This smart robot detects intruders using IoT and face recognition technology. The proposed system is equipped with passive infrared (PIR) sensor and a camera for capturing live-streaming video and photos, which are sent to the control room through IoT technology. Additionally, the system uses the face recognition algorithms to differentiate between company staff and potential intruders. The face recognition method combines high-order statistical features and evidence theory to improve facial recognition accuracy and robustness. High-order statistical features are used to capture complex patterns in facial images, enhancing discrimination between individuals. Evidence theory is employed to integrate multiple information sources, allowing for better decision-making under uncertainty. This approach effectively addresses challenges such as variations in lighting, facial expressions, and occlusions, resulting in a more reliable and accurate face recognition system. Upon detecting an unfamiliar individual, the system sends alert notifications and an email with the captured image to the control room through IoT. Additionally, a web interface was created to remotely operate the robot via a WiFi connection. The proposed method for human face recognition is evaluated, and a comparative analysis with existing techniques is conducted. Experimental results with 400 test images of 40 individuals demonstrates the effectiveness of combining various attribute images in improving human face recognition performance. Experimental results shows that the algorithm achieves an accuracy of 98.63% in identifying human faces.

Review
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Majid Joudaki

Abstract:

This literature review provides a comprehensive synthesis of research on the use of Restricted Boltzmann Machines (RBMs) and Deep Belief Networks (DBNs) for human action recognition (HAR) from 2012 to the present. The review begins by introducing the theoretical foundations of RBMs and DBNs, detailing their architectures, training algorithms (notably contrastive divergence), and various extensions—including convolutional and recurrent adaptations—that have been developed to better capture the spatial–temporal dynamics inherent in video data. Key contributions in the field are systematically analyzed, with emphasis on hybrid models that integrate RBM/DBN pretraining with modern deep learning techniques to enhance feature extraction and improve recognition accuracy. The review also examines the major benchmark datasets used in HAR research (such as KTH, HMDB51, UCF101, NTU RGB+D, and Kinetics), discussing preprocessing strategies, evaluation metrics, and the challenges associated with overfitting, computational complexity, and model interpretability. In addition, recent trends such as the incorporation of attention mechanisms, self-supervised learning, and multi-modal data fusion are explored. By highlighting both the historical significance and the evolving advancements of RBM/DBN methodologies, this review provides insights into the current state of HAR research and outlines promising directions for future investigation, including the integration of generative pretraining with emerging architectures for robust and efficient real-time action recognition.

Article
Computer Science and Mathematics
Robotics

Bojian Yu,

Petro Pavlenko

Abstract: This study aims to enhance the maneuverability and monitoring capabilities of biomimetic robotic fish in aquatic environments. A three joint biomimetic fish with forked tail fins was designed based on the body and tail fin (BCF) model to optimize propulsion efficiency and mimic natural fish swimming. The mechanical structure uses servo modules connected in series to simulate spinal joints, and precise control is achieved through PWM signals and Arduino based systems. The tail fin design follows the principles of fluid dynamics, especially the anti Karman vortex street effect, while the ballast mechanism is used to adjust the pitch angle to achieve vertical motion. The robotic fish integrates temperature, pH, and turbidity sensors for real-time water quality monitoring and data transmission through WiFi. The experimental results show that the swimming speed of the bionic fish is 0.018 m/s, the turning angular velocity is 15 °/s, and the maximum pitch angle is 5.69 °, verifying its good maneuverability. The water quality monitoring experiment shows that compared with traditional methods, the deviations of water temperature, pH value, and turbidity are 0.67%, 0.05%, and 0.23%, respectively, which meet the accuracy requirements of water quality parameter detection. This design successfully combines biomimetic motion with embedded sensing technology, providing a compact solution for dynamic aquatic monitoring. Although the system has shown potential in environmental applications, improvements in speed, maneuverability, and sensor accuracy are still needed in the future to enhance its operational capabilities in complex underwater environments.
Article
Computer Science and Mathematics
Artificial Intelligence and Machine Learning

Farhad Khosrojerdi,

Stéphane Gagnon,

Raul Valverde

Abstract: In a photovoltaic (PV) system, shading conditions caused by weather and ambient factors can significantly affect the electricity production. For more than a decade, applications of artificial intelligence (AI) techniques have been used to improve energy production efficiency in the solar energy sector. In this paper, we present how using AI-based can increase energy production for solar power plants experiencing shading conditions. It is shown that the application of these techniques paves the road towards sustainable development for the solar power sector. Employing maximum power point tracking (MPPT) control systems, running metaheuristic and computer-based algorithms, help PV arrays to cope with shading conditions effectively. Using a case study, we compare energy productions of the solar power plant in two scenarios: I) PVs without a control system, and II) PV arrays equipped with MPPT boards. System Advisory Model (SAM) is used to calculate monthly powers generated by the PV system. Our results prove our hypothesis that the PV system using MPPT systems provides a greater monthly energy production than without MPPTs.
Article
Computer Science and Mathematics
Computer Vision and Graphics

Zhongbin Luo,

Yanqiu Bi,

Qing Ye,

Yong Li,

Shaofei Wang

Abstract: Addressing the critical need for collision risk warning at unsignalized intersections, this study proposes an advanced predictive system combining YOLOv8 for object detection, Deep SORT for tracking, and Bi-LSTM networks for trajectory prediction. To adapt YOLOv8 for complex intersection scenarios, several architectural enhancements were incorporated. The RepLayer module replaced the original C2f module in the backbone, integrating large-kernel depthwise separable convolution to better capture contextual information in cluttered environments. The GIoU loss function was introduced to improve bounding box regression accuracy, mitigating issues related to missed or incorrect detections due to occlusion and overlapping objects. Furthermore, a Global Attention Mechanism (GAM) was implemented in the neck network to better learn both location and semantic information, while the ReContext gradient composition feature pyramid replaced the traditional FPN, enabling more effective multi-scale object detection. Additionally, the CSPNet structure in the neck was substituted with Res-CSP, enhancing feature fusion flexibility and improving detection performance in complex traffic conditions. For tracking, the Deep SORT algorithm was optimized with enhanced appearance feature extraction, reducing identity switches caused by occlusions and ensuring stable tracking of vehicles, pedestrians, and non-motorized vehicles. The Bi-LSTM model was employed for trajectory prediction, capturing long-range dependencies to provide accurate forecasting of future positions. Collision risk was quantified using the Predictive Collision Risk Area (PCRA) method, categorizing risks into three levels (danger, warning, and caution) based on predicted overlaps in trajectories. In the experimental setup, the dataset used for training the model consisted of 30,000 images, annotated with bounding boxes around vehicles, pedestrians, and non-motorized vehicles. Data augmentation techniques such as Mosaic, Random_perspective, Mixup, HSV adjustments, Flipud, and Fliplr were applied to enrich the dataset and improve model robustness. In real-world testing, the system was deployed as part of the G310 highway safety project, where it achieved a mean Average Precision (mAP) of over 90% for object detection. Over a one-month period, 120 warning events involving vehicles, pedestrians, and non-motorized vehicles were recorded. Manual verification of the warnings indicated a prediction accuracy of 97%, demonstrating the system’s reliability in identifying potential collisions and issuing timely warnings. This approach represents a significant advancement for enhancing safety at unsignalized intersections in urban traffic environments.

of 435

Prerpints.org logo

Preprints.org is a free preprint server supported by MDPI in Basel, Switzerland.

Subscribe

© 2025 MDPI (Basel, Switzerland) unless otherwise stated