MŽMarko M. Živanović

Marko M. Živanović

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Marko M. Živanović has a total of 21 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Hidden Variables in Real-Time Person Detection: Loss-Function Screening, Annotation Geometry, and Operating-Point Selection in YOLOX-Based Single-Class Detectors

    by Marko Živanović, Olga Ristić, Vanja Luković, Ana Savić, Hana Stefanović, Božidar Janković

    Person detection underpins a growing range of commercial systems, from retail analytics to safety monitoring, yet the experimental practices used to configure such detectors are rarely examined. This paper presents an end-to-end empirical study of single-class person detection built on YOLOX and trained on the person subset of COCO 2017 (64,115 images; 262,465 instances), stress-testing three routine practices: (i) short-run screening of bounding-box regression losses, (ii) evaluation under a fixed confidence threshold, and (iii) implicit reliance on a single annotation geometry. All three are found to be unreliable in ways that are easy to miss with a single-number comparison. Nine regression losses screened for 15 epochs on a 1000-image subset differ by only 1.06 AP, and their ranking shows no significant association with an independent benchmark (Spearman’s ρ=−0.23, p=0.55, n=9), providing no evidence that short-run screening predicts external performance. Under full 100–120-epoch schedules, the DIoU configuration reaches +1.4 AP over IoU within a fixed annotation regime—under a 20-epoch-longer schedule; so, the margin is not attributable to the loss alone—whereas switching the training and evaluation boxes from original COCO to segmentation-derived tight boxes shifts AP by up to 3 points, confounding naive cross-run comparisons; a separate SIoU+focal configuration diverges. Threshold sweeps against an off-the-shelf YOLO11m show that the better detector flips with the operating point: YOLO11m peaks at F1 = 0.870 at low thresholds, while the fine-tuned YOLOX-M retains F1 = 0.827 versus 0.751 at the common τ=0.5. Finally, hardware benchmarking (0.9 M–33.7 M parameters, 2.5–252 FPS measured directly on our hardware) and a containerized RTSP pipeline translate the findings into deployment guidance. Reporting operating-point curves, annotation geometry, and screening protocols, we argue, is as important as the loss function itself.

  • Open Access

    Comparative Analysis of CNN and Transformer Architectures for Real-Time Fire and Smoke Detection

    by Marko Živanović, Vanja Luković, Olga Ristić, Hana Stefanović, Sanja Antić, Ana Savić

    Real-time automatic detection of fire and smoke is a critical component of modern safety systems in surveillance, industrial, and environmental-protection applications, where conventional sensor-based systems show fundamental limitations. In this paper, a comparative analysis of four deep-learning architectures—DETR, Faster R-CNN (ResNet-50-FPN), Faster R-CNN (MobileNetV2), and RetinaNet—was conducted on a heterogeneous corpus of 67,765 annotated images originating from four different datasets. All models shared the same data-preparation, augmentation, and evaluation pipeline, while each was trained with the default configuration of its reference implementation; the comparison therefore reflects each architecture as it is typically deployed rather than a comparison under a single unified training budget. Faster R-CNN with the ResNet-50-FPN backbone achieved the highest accuracy (mAP@0.50 = 78.7% on the Indoor set; 73.8% on the SmokeAndFire set) and the highest mean mAP@0.50 across all datasets (48.7%), obtained with an inference speed of 76.9 FPS and a latency of 13.5 ms on NVIDIA RTX 4090 hardware, which makes it a promising candidate for real-time fire-detection systems on comparable hardware. The main contribution of this work is a unified evaluation of four representative object-detection architectures across four heterogeneous fire-and-smoke datasets. The study further quantifies the performance asymmetry between fire and smoke detection through a normalized morphological asymmetry index, reflecting the consistently lower detection accuracy achieved for smoke owing to its diffuse and semi-transparent appearance, and provides practical guidelines for selecting an architecture according to real-time deployment requirements. Because each configuration was trained once with a fixed random seed and all speed measurements were obtained on a single desktop GPU, the reported differences are interpreted descriptively; their statistical validation, cross-dataset evaluation, and measurement on embedded hardware are identified as future work.

  • Legal issues of ethical hacking: A systematic review of authorization, compliance, and liability frameworks

    by Vladimir Milićević, Miloš Jovanović, Nenad Biga, Marko M. Živanović, Igor Franc

    The practice of ethical hacking, aimed at identifying and addressing vulnerabilities in information systems, applications, or networks, presents several legal considerations. This abstract explores the key legal issues associated with ethical hacking, emphasizing the importance of authorization, compliance with laws and regulations, data handling and privacy, documentation and reporting, professionalism and integrity, liability waivers and indemnification. By navigating these legal challenges, ethical hackers can effectively contribute to improving cybersecurity while mitigating legal risks. Ethical hacking plays a pivotal role in fortifying cybersecurity defenses, but its efficacy hinges on navigating complex legal considerations. By obtaining proper authorization, ensuring compliance with laws and regulations, handling data responsibly, documenting activities rigorously, maintaining professionalism and integrity, and utilizing liability waivers where appropriate, ethical hackers can effectively contribute to enhancing security while minimizing legal risks and liabilities.

  • Open Access

    Hyperparameter optimization and validation of a deep learning instance segmentation model for real-time specific military equipment surveillance

    by Marko Živanović, Božidar Janković, Ana Savić, Nemanja Maček, Hana Stefanović

    Introduction/Objective: On the modern battlefield, a soldier's combat readiness is directly dependent on the proper use of specific military equipment (SME). Automatic verification of SME usage, primarily ballistic helmets and combat vests, is a critical factor for risk reduction and increasing combat effectiveness. Traditional object detection (bounding box) is inadequate for military conditions due to camouflage and occlusion. The objective of this paper is the development, rigorous optimization, and validation of an efficient system for real-time SME segmentation, based on a modern, lightweight YOLO12-seg architecture. Methods: The system was trained on a custom-created "Mil-SME-Seg" dataset focused on military personnel. An advanced set of data augmentations (Mosaic, MixUp, HSV-shift) was applied to achieve robustness. A systematic optimization of the key "number of epochs" hyperparameter was performed through five independent, repeatable training runs (100, 120, 150, 200, and 250 epochs) to identify the optimal balance between accuracy and overfitting. Performance was evaluated using a comprehensive set of segmentation metrics, including mAP50-95(M), mAP50(M), mAP75(M), and F1-Score(M). Results: Experiments showed that training for 150 epochs yields the best generalized results on the validation set. Peak performance was achieved at epoch 144, with an mAP50-95(M) of 0.6128, mAP50(M) of 0.7852, and an F1-Score(M) of 0.741. Metric analysis indicates that longer training runs (200 and 250 epochs) lead to clear performance degradation due to overfitting. Qualitative analysis confirms the superiority of segmentation in tactically relevant scenarios (camouflage, partial occlusion). Conclusion: The optimal number of epochs for training the model on this dataset is approximately 150. The resulting model, integrated into a system with support for dynamic "military scenarios," proved its practical applicability and tactical value, achieving real-time processing speeds (48-55 FPS). This paper lays the foundation for future systems that will use segmentation masks for equipment damage assessment and sensor fusion.

  • Temporal Pose Analysis for False Positive Reduction in Proactive Video Surveillance

    by Marko M. Živanović, Milica M. Živanović

    Proactive video surveillance demands efficient and accurate models for real-time analysis, especially in the context of protecting vulnerable groups such as the el-derly and children in public spaces. This paper presents a modular system for fall detection, a critical component of such surveillance that enables rapid emergency response. The system utilizes the YOLO11x-pose model to perform human pose estimation, processing video from diverse sources (local files, webcams, RTSP streams) to identify 17 skeletal keypoints. Its core innovation lies in detecting a fall as a dynamic transition from a standing or sitting posture to a lying state, which significantly reduces false alarms compared to static pose analysis. The methodology employs a PoseTracker for multi-person tracking and a PoseAna-lyzer that classifies posture based on biomechanical parameters (e.g., torso angle, knee angle, bounding box aspect ratio). The system generates dual outputs: a vis-ually annotated video for human review and structured JSON data for real-time integration with external alarm systems (e.g., VMS). This configurable and ro-bust solution provides a practical foundation for automated safety monitoring.

  • Development of a Model for Detecting Prompt Injection in Large Language Models Using the BERT Architecture

    by Milica M. Živanović, Marko M. Živanović

    Large Language Models (LLMs) have demonstrated remarkable proficiency in both understanding and generating natural language, which has contributed to their rapid adoption across various domains. Yet, their widespread use has also exposed them to emerging security threats, most notably prompt injection attacks. Such attacks can compromise model behavior and potentially reveal sensitive in-formation. This research explores the phenomenon of prompt injection and sur-veys existing defense mechanisms, with a particular focus on developing and evaluating detection approaches. The study formulates the detection task as a bi-nary text classification problem, distinguishing between malicious and benign prompts. Central to the analysis is the application of the BERT architecture and its lightweight variants. The main objective is to compare the performance of small-er, fine-tuned BERT-based models in identifying malicious inputs. The underly-ing hypothesis suggests that these compact models, when properly adapted, can outperform larger counterparts in detecting and mitigating injection-based attacks due to their efficiency and adaptability

  • Open Access

    A Hybrid Model Based on Chaos Theory and Artificial Immune Systems for the Analysis and Classification of Stock Market Anomalies

    by Marko M. Živanović, Emilija Kisić

    In this paper, a system for analyzing chaotic patterns in financial markets has been developed by combining classical chaos metrics with artificial immune systems for anomaly detection. Implemented indicators include the Lyapunov exponent, correlation dimension, approximate entropy, Hurst exponent, and the distance from a reference Lorenz trajectory. These metrics enable the detection of changes in market stability and predictability over time. An adaptive algorithm inspired by artificial immune systems was developed for identifying anomalous behaviors, adjusting detectors based on detected deviations. The results are presented through a series of interactive visualizations, including 3D plots, time series, and anomaly density maps. In addition to standard analysis, the system supports false alarm detection through controlled parameter variations. This approach provides deeper insights into the complex dynamics of financial markets and can serve as a tool for forecasting periods of instability.

  • Open Access

    Mathematical decomposition of prompt engineering in Large Language Model architecture

    by Miloš Jovanović, Marko M. Živanović, Aca Aleksić

    Large Language Models (LLMs) represent the convergence of neural language processing and high-dimensional statistical inference. Despite their impressive capabilities, these systems remain inherently probabilistic, generating outputs via autoregressive sampling from learned distributions. The resulting stochastic nature manifests through phenomena such as hallucinations and semantic decomposition. This paper formalizes the mathematical framework of prompt engineering as a methodology for topological navigation through the model's latent space. Through a rigorous analysis of the Transformer architecture, multi-head attention mechanisms, positional encoding, and loss functions, we deconstruct how precisely constructed prompts manipulate probability distributions during autoregressive generation. We present a formal taxonomy of ten advanced techniques - including Chain-of-Verification (CoVe), Constitutional AI, and Meta-Prompting - and demonstrate their effect on reducing the entropy of output distributions. Experimental results indicate that the systemic application of these techniques can transform a model with a baseline accuracy of 62.3% into a system with 91.7% accuracy, effectively converting a stochastic generator into a quasi-deterministic reasoning engine.

  • Primena federativnog učenja u predikciji aktivnosti studenata na e-learning kursevima

    by Fakultet Informacionih Tehnologija Univerzitet Metropolitan, Marko Živanović, Danijela Milošević, Fakultet Tehničkih Nauka Univerzitet U Kragujevcu

    Ovaj rad istražuje primenu federativnog učenja (Federated Learning, FL) u predikciji poseta kursevima na osnovu log zapisa interakcija studenata. Federativno učenje omogućava treniranje modela na distribuiranim podacima bez potrebe za njihovim centralizovanjem, čime se poboljšava privatnost i sigurnost podataka. Istraživanje uključuje tri klijenta koji predstavljaju različite kurseve (Veštačka inteligencija, Operativni sistemi i Računarski hardver) i koristi LSTM neuronske mreže za vremensku predikciju broja poseta. Modeli se treniraju lokalno na svakom klijentu, dok se agregacija modela vrši na centralnom serveru koristeći Federated Averaging (FedAvg) algoritam. Performanse FL modela upoređene su sa centralizovanim modelima koristeći metriku modela R² Score. Rezultati pokazuju da FL pristup može postići konkurentne rezultate u poređenju sa centralizovanim modelima, uz značajne prednosti u zaštiti privatnosti podataka. Ovaj rad pruža doprinos u oblasti analitike učenja i otvara mogućnosti za dalja istraživanja primene FL u obrazovnim sistemima.


  • Model rasplinute evaluacije kvaliteta softverskog sistema zasnovan na fuzzy logici

    by Fakultet Informacionih Tehnologija Univerzitet Metropolitan, Marko M. Živanović, Olga Ristić, Fakultet Tehničkih Nauka Univerzitet U Kragujevcu

    U ovom radu predstavljena je metodologija za evaluaciju kvaliteta veb sajtova akademskih i državnih institucija iz svih šest republika bivše SFRJ korišćenjem alata iz oblasti veštačke inteligencije, konkretno fuzzy logike. Razvijen je sistem za automatsku procenu performansi sajtova na osnovu tri ključna parametra: vreme učitavanja stranice, veličina stranice i broj HTTP zahteva. Na osnovu ovih ulaznih vrednosti, fuzzy kontrolni sistem generiše ocenu kvaliteta sajta na skali od 0 do 10, uz proces defazifikacije. U evaluaciju su uključeni sajtovi univerziteta, ministarstava, naučnih instituta i agencija, čime se pruža sveobuhvatan uvid u tehnički kvalitet javno dostupnih digitalnih resursa u regionu. Pored numeričke ocene, sistem omogućava i vizualizaciju pripadnosti ulaznih vrednosti fuzzy skupovima, čime se dodatno objašnjavaju uticaji pojedinačnih faktora na ukupnu ocenu. Rezultati ukazuju na značajne razlike u tehničkoj optimizaciji analiziranih sajtova, čime se otvara prostor za buduća unapređenja u dizajnu i implementaciji veb prezentacija institucija od javnog značaja.