Abstract: Cable theft is a major infrastructure security issue in South Africa, especially in Gauteng municipalities, where poverty is spatially concentrated to the extent that it has given rise to large-scale theft of electricity distribution infrastructure components. This essay focuses on the complexity of the reasons behind cable theft, including unemployment-related desperation, organized crime syndicates, governmental corruption, and international commodity markets. The study involved 19,919 reported cases from around the country and 5,914 cases from Gauteng since April 2019. It is found that cable theft is not limited to property crime, but also poses a threat to basic service delivery in the electricity, water, health, and education systems. The study reveals that effective mitigation should involve combined interventions using advanced technology (intrusion detection systems, smart meters, and mechanical barriers), enhanced law enforcement and prosecution, community involvement, and socioeconomic development measures. Evidence-based policies, such as special detection offices, scrapyard control, dedicated prosecution, and legitimacy programs that integrate infrastructure security and affordability approaches, are proposed as critical elements of sustainable reduction strategies. This study concludes that the solution to cable theft is closely connected to the constitutional obligation of South Africa to progressively fulfil socioeconomic rights; thus, municipal, provincial, and national governments must act in collaboration to address this issue.

Abstract: Over-the-air (OTA) update systems are used to deliver software in real time for fields such as aviation, railway systems, and medical devices. This study builds a cloud-based OTA setup using Kubernetes and Istio to improve update speed and system stability across different types of devices. The system includes rolling updates, blue-green switching, gRPC transmission, and message queue scheduling. Tests were run on 72 terminals from vehicle, avionics, and medical settings. Results show that the average image transfer time dropped from 842 ms to 493 ms, and the failure rate was reduced from 3.6% to 0.8%. In 500 failure tests, the average time to restore service became 38.7% shorter. These results confirm that using containers and service-level routing helps shorten delays and reduce errors in OTA processes. The method can be applied in real-world embedded systems but may require extra tuning on older hardware or unstable networks.

Abstract: Background: Cognitive aging is highly heterogeneous, not only in performance but also in how individuals perceive their own aging. Such self-perceptions may shape emotional reactions and adaptation to memory difficulties, yet little is known about their organization in patients referred to a memory clinic for a first diagnostic consultation. The primary aim of this study was to identify the internal configuration of self-perceptions of aging in such patients. A secondary aim was to compare these patterns with those observed in older adults recruited in a research unit of experimental psychology, who reported subjective complaints but had no medical referral. Methods: In total, 130 memory clinic patients and 84 laboratory participants completed, prior to the same neuropsychological testing, a psychosocial questionnaire assessing four domains: self-perceptions of memory deficits, attitudes toward aging, aging stereotypes, and multiple facets of subjective age. Network analysis was applied to examine how these variables were interrelated and to determine their relative importance in each group. Results: Across both samples, network analyses revealed distinct organizational patterns. Patients showed a unified representational system characterized by more positive associations and centered on subjective age variables. By contrast, the laboratory group showed a two-cluster network with more negative connections, organized around negative aging stereotypes. Conclusions: These findings provide novel insights into the psychosocial profile of memory clinic patients, highlighting the central integrative role of subjective age in integrating emotional responses, aging beliefs and perceptions of memory difficulties, and underline the value of network approaches in capturing heterogeneity in cognitive aging.

Abstract: It is still an open question that how the masses are formed for charged leptons. The widely accepted Yukawa coupling mechanism generally have quite randomized coupling constants for their masses. In this paper, we tried to build a simple model to calculate the masses of charged leptons. We assumed that the masses are formed by coupling of the plasma characteristic energy from the particle-antiparticle pairs in the background sea, and the electric potential inside the Compton ball. The internal structure of the charged leptons is thought to have three states, i.e. the negative charge, the positive charge and the Planck scale Kerr black hole. For electron and muon, the zitterbewegung is formed by positive charge and negative charge, but for tau, the excited zitterbewegung is formed by negative charge and the Planck scale Kerr black hole. The calculations of this simple model give quite close values for the charged leptons as compared with the lab results. We think we may need pay more attention on the internal structure of a particle.

Abstract: Convolutional neural networks are increasingly used for photovoltaic fault recognition from RGB imagery, yet high benchmark accuracy can mask shortcut learning induced by heterogeneous backgrounds, viewpoints and class imbalance. Using the Kaggle "PV Panel Defect Dataset" dataset, we compare five architectures (Baseline CNN, VGG16, ResNet50, InceptionV3 and EfficientNetB0) through a complementary explainability pipeline: LIME superpixel surrogates (with kernel-weighted R2 fidelity), occlusion sensitivity (functional relevance under localized masking) and Integrated Gradients (IG) validated by deletion-insertion curves. To reduce reliance on subjective saliency inspection, we quantify localization and concentration using IoU@Top10% against consistent proxy defect masks, Shannon entropy and Hoyer sparsity, and we summarize IG faithfulness with a Faithfulness Gap (AUC_insertion - AUC_deletion) and an accuracy-faithfulness consistency score at class level. ResNet50 attains the best predictive performance (82.3% accuracy), while EfficientNetB0 provides the strongest overall evidence faithfulness (mean Faithfulness Gap ~ 0.019) and stable, panel-centered attributions. InceptionV3 frequently yields diffuse relevance, and VGG16 produces highly concentrated but occasionally brittle hotspots. Bird-drop and Snow-covered show the most consistent alignment between accuracy and faithful evidence, whereas Clean and the two damage classes remain vulnerable to context cues (e.g., borders and background textures). The results support integrating quantitative explainability diagnostics into PV model selection and dataset curation to mitigate shortcuts and improve trustworthiness in vision-based PV monitoring.

Abstract: This research introduces an edge-optimized reinforcement learning (RL) ecosystem engineered for sustainable logistics in the blue economy, spanning maritime shipping, automated port operations, and offshore resource transportation. At its core, the system processes vast streams of real-time data from IoT sensors embedded in vessels, buoys, and drones directly at edge nodes, bypassing cloud latency to enable instantaneous decision-making in unpredictable marine conditions like storms or currents. Carbon capture analytics, derived from spectroscopic sensors quantifying direct air capture (DAC) efficiency and CO2 sequestration rates on ships, dynamically adjusts RL reward functions to favour fuel-efficient paths that maximize emissions offsets, aligning with International Maritime Organization (IMO) mandates for net-zero operations by 2050. The framework exploits 6G networks' terabit speeds, sub-millisecond latency, and non-terrestrial network integration via low-earth-orbit satellites for seamless swarm intelligence orchestration. Autonomous agents unmanned surface vessels (USVs), aerial drones, and autonomous underwater vehicles (AUVs) exhibit flocking behaviour’s inspired by particle swarm optimization, sharing pheromone-like digital signals over holographic beamforming channels to collaboratively resolve complex tasks like dynamic routing, collision avoidance, and load redistribution. Methodologically, proximal policy optimization (PPO) algorithms facilitate stable, lightweight training on resource-constrained edge hardware, augmented by federated learning to aggregate insights across privacy-sensitive multi-operator fleets without central data pooling. Rigorous evaluations in NS-3 for 6G emulation and Gazebo for maritime physics reveal transformative gains: 42% reductions in carbon footprints, 65% lower end-to-end latency versus 5G-cloud hybrids, and 30% improvements in throughput under adverse weather. Scalability tests with 1000+ agents confirm robustness in GPS-denied zones, while ablation studies highlight the synergistic impact of carbon feedback and swarm coordination over siloed baselines like genetic algorithms or centralized RL. By embedding quantum-safe encryption for 6G links and digital twin interfaces for predictive maintenance, this ecosystem not only decarbonizes blue economy logistics but also sets a scalable blueprint for AI-driven sustainability in cyber-physical systems worldwide.

Abstract: Over-the-air (OTA) updates in multi-tenant systems often face task conflicts, cache overlap, and weak fault recovery during parallel updates. This study designed a layered fault-tolerance and isolation method that combines task redundancy, cache separation, and snapshot rollback. Tests were carried out on 120 devices across six tenants with a fault rate of up to 95%. The system kept stable operation, extended the mean time between failures (MTBF) to 182 hours, and raised total availability from 98.2% to 99.7%. The average update delay per tenant stayed below 1.1 seconds, showing that higher reliability did not slow the process. The method effectively avoided tenant interference, reduced recovery time, and improved update stability. It provides a simple and practical solution for dependable OTA updates in industrial, automotive, and IoT systems.

Abstract: The use of light (photons) delivered clinically from laser or light-emitting diodes (LED), is referred to as photobiomodulation therapy (PBMt). Increasingly PBMt is accepted particularly in dental practice for pain or pre-emptive anaesthesia. Understanding its mechanism of effectiveness is the key to its increasing acceptance. Of major importance to this is how PBMt affects not only the neurons but also the Schwann cells and fibroblasts of the peripheral nervous system which are unique in morphology and function. The specific roles of the neuronal cells of the dorsal root and trigeminal ganglia, now include consideration of the axon initial segment responsible for the initiation of the action potential and the T junction from which the distal and proximal axons arise which are complex but central to normal function. This cellular complexity, organization and function is discussed leading to a review of the mechanism of effectiveness of PBMt demonstrated by clinical trials in both medicine and dentistry. This review provides evidence of the involvement of the cytoskeleton, mitochondrial organization particularly related to fast and slow axonal flow and mitochondrial membrane potential in response to light in somatosensory neurons and nerves.

Abstract: The centrosome, long recognized as the primary microtubule-organizing center (MTOC) of animal cells, is re-examined through the lens of information theory and systems biology. This preprint proposes a unifying hypothesis: the mother centriole within the centrosome acts as a non-genetic cellular ledger, a stable structural repository that accumulates molecular records of a cell’s replicative history and environmental exposures. These records—comprising specific post-translational modification (PTM) signatures, retained proteins, and structural alterations—are subsequently “read” by the cell to inform critical decisions regarding proliferation, differentiation, senescence, and apoptosis. We synthesize evidence from cell biology, gerontology, and evolutionary biology to construct the “Centrosomal Ledger Model.” This model positions the centriole not as a passive cytoskeletal component but as an active, heritable information-processing node that integrates temporal data across scales—from circadian rhythms to organismal aging. We detail the molecular mechanisms of information encoding (e.g., tubulin polyglutamylation, oxidative marks) and decoding (via ciliary signaling, proteostatic feedback, and mechanical transduction). The model’s implications challenge genetic determinism by highlighting structural inheritance, provides a material basis for cellular age, and offers novel, falsifiable avenues for experimental interrogation in aging and cancer research. Crucially, it suggests that modulating the “read-write” cycle of the centrosomal ledger could represent a new frontier in regenerative medicine.

Abstract: This study assessed the spatial distribution and composition of airborne particulate matter within a 10-km long urban green corridor in Zacatecas, Mexico, using magnetic biomonitoring with Tillandsia recurvata and SEM-EDS particle characterization. A total of 44 samples were collected from distinct urban park contexts (e.g., commercial zones, malls, bus stop), revealing mass-specific magnetic susceptibility χ values ranging from -6.71 to 61.1 × 10⁻⁸ m³kg-1. Three compositional groups were identified based on a PCA performed using elemental concentrations from SEM-EDS and magnetic data, which are associated with traffic emissions and industrial inputs. SEM-EDS images confirmed abundant magnetite-like particles (1–8 μm) with hazardous metals including Pb (up to 5.6 wt.%), Ba (up to 67.6 wt.%), and Cr (up to 31.5 wt.%). Wind direction data indicated predominant SSW-NNE transport, correlating with hotspots in central and northeast-ern park areas. Overall, vegetated zones displayed significantly lower magnetic loads (mean χ = 8.84 × 10⁻⁸ m³kg⁻¹, σ = 6.65 × 10⁻⁸ m³kg⁻¹) compared to traffic-exposed sites (mean χ = 17.27 × 10⁻⁸ m³kg⁻¹, σ = 12.44 × 10⁻⁸ m³kg⁻¹), emphasizing the pollution mitiga-tion role of green barriers. This research highlights the applicability of combined mag-netic and microscopic techniques for evaluating the dynamics of airborne pollution in urban parks and supports their use as biofunctional filters in cities facing vehicular air pollution.

Abstract: This paper undertakes a foundational exploration of the nature of mathematics from both historical and philosophical perspectives, with a primary focus on the Indian intellectual tradition. It traces the evolution of mathematical thought from ancient Vedic texts such as the ´Sulba S¯utras, through the formal grammar of P¯an. ini, to modern abstract mathematics including group theory, automata, and topology. The investigation is rooted in the dual inquiries of ontology and epistemology, examining what it means for mathematics to be and how mathematical knowledge is constructed and validated. Particular emphasis is placed on the Indian concepts of gan. ita (mathematics), ´s¯unya (zero), and ´s¯unyat¯a (emptiness), and their correspondence with Western notions such as the Cartesian dualism, the set-theoretic empty set, and symbolic logic. The paper explores the recursive cosmological cycles found in Indian time theory, mathematical cosmology, and ritual geometry, showing how these ideas anticipated or paralleled developments in modern mathematics, including measure theory, combinatorics, and fractals. With detailed references to logical systems (Ny¯aya), sacred architecture (v¯astu-´s¯astra), cyclic time constructs (kalpas and yugas), and formal structures in linguistic grammar (As.t. ¯adhy¯ay¯ı), the paper argues for a view of mathematics as both a sacred science and a system of abstract formalism. Across these investigations, mathematical structures are treated not merely as tools for calculation but as profound reflections of metaphysical principles, visualizable through mandalas, yantras, and cosmological diagrams. This study invites a reassessment of how different cultures have understood and visualized mathematics as an expression of cosmic and cognitive order.

Abstract:

Leuenbergeria bleo (Kunth) DC. (Cactaceae), previously classified as Pereskia bleo, represents a phylogenetically basal cactus species with a disjunct distribution across Central America, Southeast Asia, and southern China. Phytochemical investigations have traditionally emphasized small-molecule secondary metabolites, including phenolics, alkaloids, and terpenoids, which contribute to antioxidant and anti-inflammatory activities. However, recent peptidomic analyses have expanded this chemical space through the discovery of bleogens, a family of hyper-stable, cysteine-rich microproteins with specific antifungal and wound-healing properties. This review systematically integrates botanical characteristics, ethnomedicinal applications, and pharmacological profiles, providing a comparative analysis of the plant’s small-molecule constituents versus its peptidyl biologics. It identifies the co-existence of these distinct chemical classes as a defining feature of the plant’s efficacy while highlighting the need for future research into their potential interactions.

Abstract: The increasing demand for recombinant proteins has driven innovation in bioprocessing strategies using Komagataella phaffii as a host organism. Conventional fed-batch cultivation with pure methanol induction remains widely used but presents challenges including high methanol consumption, extended downtime, and elevated operational costs. This study evaluates alternative strategies combining mixed induction (methanol/sorbitol) with continuous cultivation to enhance productivity, sustainability, and improved economic outcome. Using KEX2 protease as a model industrial recombinant protein, we compared four cultivation modes: fed-batch with methanol (benchmark), fed-batch with mixed induction, continuous with methanol, and continuous with mixed induction. Cell growth, volumetric yield, and specific productivity were evaluated at 5L scale and then modelled to simulate industrial scales (40 L and 400 L). Results demonstrate that continuous cultivation with mixed induction significantly improves yield up to 9-fold compared to conventional fed-batch and reduces methanol usage and oxygen demand. Techno-economic simulations reveal that a 40 L continuous process can match or exceed the output of two 400 L fed-batch runs, while lowering capital and operating costs and minimising environmental footprint. This integrated strategy offers a scalable, cost-effective, and safer alternative for recombinant protein production, supporting the development of compact and sustainable manufacturing platform

Abstract:

The Warburg effect, classically defined as the preferential use of glycolysis by cancer cells in the presence of oxygen, has been a central concept in cancer biology since a long time. Otto Warburg had originally proposed that defective mitochondrial respiration was the primary cause of aerobic glycolysis in cancer cells. While this hypothesis profoundly influenced early cancer metabolism research, it has now become increasingly clear that this interpretation has gaping. Advances in biochemistry, molecular biology and metabolomics demonstrate that mitochondria in many cancers are functional and play essential roles in biosynthesis, signaling and energy production. Aerobic glycolysis in cancer cells is now recognized as an adaptive metabolic strategy that supports rapid proliferation by providing metabolic intermediates, maintaining redox balance, and enabling cellular signaling rather than maximizing ATP yield. This review discusses the Warburg effect through the lens of modern cancer metabolism. It contrasts classical misconceptions with current evidences, discusses key regulatory pathways like HIF-1α, PI3K/Akt/mTOR, c-Myc and PKM2, and examine the central role of lactate as both a metabolic fuel and a signaling molecule. It further explores metabolic heterogeneity, the reverse Warburg effect, immune–metabolic interactions, and the relevance of oxidative phosphorylation in cancer. Finally, some unresolved questions are highlighted that is critical for future understanding of cancer metabolism.

Abstract: This systematic review examined the validity and reliability of wearable inertial sensor systems to quantify spatiotemporal gait parameters in post‑stroke adults, a population in which gait asymmetry, altered motor control, and compensatory strategies challenge accurate measurements. Four databases were searched up to December 2025, and studies were included when they assessed concurrent validity or test–retest reliability of wearable derived spatiotemporal parameters against laboratory-based reference systems. Fifteen studies involving a total of 286 participants were analyzed. Spatial parameters as gait speed, cadence, and step and stride length showed consistent good‑to‑excellent agreement with reference instruments and high test–retest reliability. Temporal parameters demonstrated greater heterogeneity, with larger absolute errors, wider limits of agreement, and lower relative agreement, particularly for swing time. Paretic‑side measurements showed reduced between instrument agreement compared to the non‑paretic side, although within‑subject reliability remained moderate to high. No consistent influence of sensor number on measurement accuracy. Overall, wearable inertial sensors provide robust estimates of spatial gait parameters in post‑stroke populations, while temporal outcomes remain limited, likely due to the challenge to detect gait events that arise from stroke-related alterations in gait biomechanics. Taking these findings as a whole suggests that deviations from regular gait biomechanics, whether due to reduced speed particularly at low walking speeds of 0.4 m/s, asymmetry, or to the use of assistive devices, compromise the ability of wearable-based algorithms to accurately identify gait events.

Abstract: This paper presents a transformer-infused semantic sensing ecosystem that integrates post-quantum signatures with 6G-enabled digital twins to enable adaptive orchestration in next-generation smart systems. Conventional IoT architectures struggle with semantic understanding across heterogeneous sensor streams, vulnerability to quantum attacks, and synchronization delays between physical and digital representations. The proposed platform deploys transformer models optimized for multi-modal sensor fusion to extract contextually rich semantic features from raw measurements, feeding these insights into digital twins synchronized over 6G networks with microsecond precision. Post-quantum lattice-based signatures ensure data integrity and authentication across the high-velocity sensing-orchestration pipeline, resisting both classical and quantum adversaries. The adaptive orchestration engine leverages twin predictions and semantic context to generate control policies that optimize system performance under dynamic conditions. Evaluation across industrial, urban, and autonomous transport scenarios demonstrates 3.8× improvement in semantic inference accuracy, 92% reduction in twin synchronization error, and 28% latency reduction compared to baseline architectures, while maintaining quantum-resistant security guarantees. The framework establishes a blueprint for secure, semantically-aware smart ecosystems capable of real-time adaptive orchestration at 6G scale.

Abstract: Central tolerance is provided by the AIRE-expressing medullary thymic epithelial cells, through high avidity recognition of self-antigens. Nevertheless, peripheral mechanisms regulate adaptive immunity by deleting autoreactive T-cells that escape thymic selection or inducing their functional unresponsiveness. These mechanisms require interaction with antigen presenting cells exposing cognate antigen. As regard multiple types of extrathymic AIRE-expressing cells, residing in secondary lymphoid organs, were described. In this study we aimed to provide evidence for AIRE binding to promoter regions of known autoantigens in human peripheral blood mononuclear cells (PBMC) in an attempt to elucidate whether this non-classical transcriptional factor could play a role in the pe-ripheral expression of self-antigens. Chromatin immunoprecipitation (ChIP) of 4 normal human PBMC samples was performed using anti-AIRE monoclonal antibody. Quantitative-Real-Time PCR (qRT-PCR) was used to detect AIRE-binding at promoters of known autoantigens, including thyroidrelated thyroglobulin, thyroperoxidase, thyrotropin-receptor, Type 1 diabetes-related autoantigens, i.e. insulin and zinc transporter 8, and to confirm their expression in PBMC. ChIP evidenced amplicons of promoter regions of mentioned autoantigens by qRT-PCR. Expression of AIRE and of autoantigens was confirmed in the same human PBMC samples. This study provides the first evidence that AIRE binds promoters of known autoantigens in human PBMC, supporting its expression and potential role in modulating peripheral self-antigen expression.

Abstract: Background: Artificial intelligence (AI) shows promising results in lymphoma detection, prediction, and classification. However, translating these findings into practice requires a rigorous assessment of potential biases, clinical utility, and further validation of research models. Objective: The goal of this study was to summarize existing studies on artifi- cial intelligence models for the histopathological detection of lymphoma. Design: This study adhered to the PRISMA Extension for Scoping Reviews (PRISMA-ScR) guidelines. A systematic search was conducted across three major databases (Scopus, PubMed, Web of Science) for English-language articles and reviews published between 2016 and 2025. Seven precise search queries were applied to identify relevant publications, accounting for variations in study modality, algorithmic architectures, and disease-specific terminology. Results: The search identified 615 records, of which 36 articles met the inclusion criteria. These studies presented 36 AI models, comprising 30 diagnostic and 6 prognostic applica- tions, with Convolutional Neural Networks (CNNs) being the predominant architecture. Regarding data sources, 83% (30/36) of datasets utilized Hematoxylin and Eosin (H&E) stained images, while the remainder relied on diverse modalities, including IHC stained slides, bone marrow smears, and other tissue preparations. Studies predominantly utilized retrospective, private cohorts with sample sizes typically ranging from 50 to 400 patients; only a minority leveraged open-access repositories (e.g., Kaggle, TCGA). The primary appli- cation was slide-level multi-class classification, distinguishing between specific lymphoma subtypes and non-neoplastic controls. Beyond diagnosis, a subset of studies explored advanced prognostic tasks, such as predicting chemotherapy response and disease progres- sion (e.g., in CLL), as well as automated biomarker quantification (c-MYC, BCL2, PD-L1). Reported diagnostic performance was generally high, with accuracy ranging from 60% to 100% (clustering around 90%) and AUC values spanning 0.70 to 0.99 (predominantly >0.90). Conclusions: While AI models demonstrate high diagnostic accuracy, their translation into practice is limited by unstandardized protocols, morphological complexity, and the "black box" nature of algorithms. Critical issues regarding data provenance, image noise, and lack of representativeness raise risks of systematic bias, hence the need for rigorous validation in diverse clinical environments.

Abstract: Background: persistent strength deficits and psychological impairments may compro-mise return to sport (RTS) after anterior cruciate ligament reconstruction (ACLR). Ob-jective: to investigate the relationship between thigh muscle isokinetic strength recovery at six months after ACLR and long-term psychological outcomes related to RTS in competitive male soccer players. Methods: sixty male soccer players who underwent primary ACLR with bone–patellar tendon–bone autograft were retrospectively analyzed. Isokinetic testing of quadriceps and hamstrings was performed one week before surgery and six months post-surgery at 90°/s and 180°/s. Limb symmetry index (LSI) was calcu-lated both pre- and post-operatively. At long-term follow-up (mean ≈4 years after RTS), athletes completed questionnaires assessing RTS status, ACL re-injuries, sport-related perceptions, and kinesiophobia using the Tampa Scale for Kinesiophobia (TSK). Results: absolute quadriceps and hamstring peak torque values significantly increased from pre- to post-surgery, with quadriceps strength deficits persisting only in the operated limb. However, quadriceps LSI significantly decreased post-operatively, while hamstring LSI remained stable. Overall, RTS rate was 91.7%, but a second ACL injury occurred in 18.2% of athletes. High kinesiophobia (TSK ≥ 37) was present in 56.7% of the cohort at long-term follow-up. Conclusions: despite significant strength gains, quadriceps limb symmetry worsened six months after ACLR, with deficits confined to the operated limb, suggesting persistent neuromuscular inhibition. These physical deficits coexist with long-term ki-nesiophobia despite high RTS rates. Findings highlight the limitations of time-based and strength-only RTS criteria and support the need for an integrated physical, psychological, and neuro-cognitive approach to rehabilitation and RTS decision-making.

Abstract: Accurate prediction of pedestrian intention and future paths is essential for traffic safety, urban planning, and autonomous navigation. This study develops a multimodal prediction model that combines meaning-based image-text features, motion trajectories, and social interactions. We extract visual-language information from RGB sequences using a CLIP-based encoder and represent group behavior using a Social-GRU network. To improve the reliability of predictions, we apply Bayesian modeling to manage uncertainty. We tested the method on the Waymo and ETH/UCY datasets. On the ETH dataset, the model achieved a 14.2% reduction in average displacement error and a 17.6% reduction in final displacement error, compared with leading baseline methods. The model remained effective in crowded spaces, unclear visual conditions, and sudden motion changes. The results confirm that combining visual-language and motion data improves prediction accuracy. This method offers a practical solution for real-world pedestrian analysis in intelligent transport systems.