1. Introduction
The Special Issue is focused on object detection and recognition, a central task in almost all applications of computer vision, showing the capability of novel mathematical algorithms, architectures and methods to improve the performance in this task, with emphasis placed on new solutions and analysis of their performance in challenging conditions in relevant applications.
The computer vision community has adopted deep-learning models in the last decade due to their superior performance with respect to classical methods. These models require high processing power (GPUs) for training with large datasets and provide inferences in real time, and have been dominated by convolutional neural networks (CNNs). They are subdivided in two types: two-shot detectors, searching maximum accuracy with potential cost of inference time, and one-shot detectors, oriented to minimum inference time for real-time applications. Two-shot detectors are dominated by the R-CNN family (region-proposal CNNs), such as Fast R-CNN, Faster R-CNN or Cascade R-CNN solutions, while the YOLO family dominates for one-shot detectors, which also includes SSD and RetinaNet as the most popular algorithms in this category. Additionally, in recent years, the family of Vision Transformers (ViTs) has also been applied to object detection and recognition tasks. ViT-based algorithms, such as DETR or DINO, are based on a self-attention mechanism, which learns the relationships between elements of a sequence, applying the transformer architecture on image grids. ViTs make use of CNNs as a backbone for feature extraction, given their strong ability to automatically extract relevant features. In addition, object detection is closely related with other open challenges in machine vision as Multi-Object Tracking (MOT), which involves both detection and tracking of objects of interest appearing in the video sequence. The goal in this case is to not only identify and locate in each frame but also associate them across frames to keep track of continuity and follow their dynamics over time.
Regarding evaluation, it is complex developing fair comparisons among different solutions, considering the balance between accuracy and speed, resolution of input images, configuration of evaluation parameters, etc. Additionally, the conditions may differ across different situations, making it difficult to achieve general solutions that are robust against domain shift, usually requiring further fine tuning when applied to new scenarios even with minimal differences with respect to training scenarios. It is typical to make the analysis based on available benchmarks and datasets, which are necessary to evaluate the performance of different architectures and configurations. In this sense, many authors have identified class imbalance as an additional challenge to achieve high accuracy. In this sense, other deep-learning architectures, such as GAN or autoencoders, can be combined with detectors to enhance the training phase, increasing the sizes and variety of the datasets, such as to improve the detection of very small objects. Additionally, learning can be improved for imbalanced situations by adapting the loss function to focus learning on hard examples and avoid bias towards numerous negative examples.
2. Contributions
This Special Issue collects 12 recent works on the areas indicated above. They are briefly summarized in this section, although readers are encouraged to read them in detail and consult the bibliography included in each one.
The first contribution proposes a monocular 3D multi-object tracking solution, D3SSTrack, based on a tracking-by-detection paradigm. It integrates a Mamba state-space model extended with dropout regularization and an additional projection layer to improve feature extraction before temporal filtering. This enriched backbone with a state-space tracker learns temporal consistency and preserves identity embeddings for a more robust data association process. The second contribution describes CB-OWL-ViT (a Cluster-Based Open-World Localization Vision Transformer), an adaptable monitoring system to support public health measures such as mask compliance and social distancing, developed for situations like COVID-19. It incorporates homography-based distance estimation with monocular cameras, combined with an open-world vision-language detection and a clustering-based strategy to enhance mask-wearing detection without retraining. Detection and social distance estimation are performed independently for each frame, leaving the application of tracking modules such as ByteTrack or DeepSORT for future works.
In the third work, a framework for image restoration is proposed, also in masking situations, named M2UNet (Masked-to-Unmasked Network). Current inpainting methods often struggle with structural fidelity when handling large-area occlusions, leading to blurred or inconsistent results. M2UNet is a segmentation-guided generative framework, using a segmentation-derived mask prior to accurately localize occluded regions and employs a multi-scale, attention-enhanced generator to restore fine-grained facial textures. It is evaluated on the synthetically masked CelebA dataset, which allows quantitative evaluation with ground truth, leaving the collecting and testing on a dataset of real-world masked faces for a future work. Next, the fourth contribution in the issue introduces KORIE, a curated benchmark of 748 Korean retail receipts designed to evaluate scene text detection, Optical Character Recognition (OCR), and Information Extraction (IE) under challenging digitization conditions. It includes exclusively receipts digitized via flatbed scanning (HP LaserJet MFP), specifically selected to preserve complex thermal printing artifacts such as ink fading, banding, and mechanical creases. Authors compare fully supervised models in the state of the art for detection (YOLOv9, YOLOv10, YOLOv11, and DINO-DETR), OCR, (EasyOCR, PaddleOCR), and Information Extraction of zero-shot Large Language Models (Llama-3, Qwen-2.5).
The fifth contribution presents MSLCP-DETR, an enhanced RT-DETR-based framework that integrates multi-scale linear attention and sparse fusion mechanisms for detecting small infrared targets from vehicle-mounted systems. It combines multi-branch convolution with linear attention (MSLA-AIFI) to capture cross-scale features while reducing computational complexity; a Cross-Scale Small Object Feature Optimization module (CSOFO) to enhance the localization of small targets in dense scenes through spatial rearrangement and dynamic modeling; and a Pyramid Sparse Transformer (PST) to replace traditional dense fusion with a dual-branch sparse attention mechanism to improve accuracy and real-time performance. Experimental results demonstrate improvement of MSLCP-DETR over the baseline RT-DETR-R18 in 2.7%, achieving superior performance in both parameter count and computational complexity. Then, the sixth work proposes a feature point extraction method named Large Viewpoint Feature Extraction (LV-FeatEx) for images under large viewpoint changes. The challenge is maintaining stable image feature extraction when the angle between the camera’s reverse direction and the object’s surface normal exceeds 40 degrees. It employs optimized affine simulation to achieve perspective-distortion-aligned feature point extraction, enhancing stability under large viewpoints compared with classic feature point extraction and deep learning-based methods. To achieve high performance in both accuracy and speed, it uses a dual-threshold approach to constrain the AGAST (Adaptive and Generic Accelerated Segment Test) feature detector combined with the FREAK (Fast Retina Keypoint) descriptor to enable more effective estimation of camera motion parameters.
The seventh contribution addresses Automatic Meter Reading (AMR) technologies for utilities in the public services such as water, electricity or gas, by allowing users to regularly upload images of their meters to be automatically processed for digit recognition. It proposes an approach designed for real environments with failures such as image blur, perspective distortion, partial reflection, poor lighting, missing digits or intermediate digit states. The proposal first checks the quality of the user-uploaded images through the SVM method and requests re-uploads for images unsuitable for digit extraction and recognition. The second stage employs deep learning models for digit localization and recognition, automatically detecting and correcting issues such as missing and intermediate digits to enhance the accuracy of automatic meter readings. The eighth contribution presents an investigation of motorbike detection on edge devices for real-time applications, considering different embedded edge devices and SSD Mobilenet and YoloV5 detectors, highlighting the difference conventional machine learning implementation and edge deployable machine learning requiring optimization of the neural network. Model size is significantly reduced by several folds after optimization to reduce computation complexity and increase inference rate at the cost of some decrease in mAP. This decrement lies in the range of 5 percent of the original baseline accuracy, while the accuracy of motorbike detection is improved with a higher FPS. The ninth contribution explores the use of a multimodal Contrastive Language-Image Pretraining (CLIP) model on a multiclass dataset to obtain reasonable results with few images per class. It is applied in the domain of human posture classification in images and videos, exploring CLIP as a robust fine-tuner, apart from its usual capability for understanding and interpreting images in the context of natural language, showing improvements with respect to the previous results in a dataset of yoga pose classification model.
The tenth work in the issue addresses fire detection within enclosed spaces, covering the detection of both fire instances and combustible materials. It presents a novel dataset collected of fire and materials, the Material-Auxiliary Fire Dataset (MAFD), and the analysis of deep learning techniques to achieve both fire and material recognition and segmentation in the indoor scene. A Dual Attention Network (DANet) was specifically designed for image semantic segmentation to recognize fire and combustible material, outperforming the prevalent methods applied in this domain (PSPNet, CCNet, FCN, ISANet, OCRNet). The eleventh work addresses functional movement screening (FMS), a testing method to identify functional limitations in the human body by inspecting fundamental movement patterns. It introduces an automatic assessment approach for FMS based on an Inflated 3D ConvNet network (I3D) to extract spatiotemporal video features across various scales and levels, followed by an attention mechanism (AM) module to enable the network to focus on human movement characteristics, and then by a multilayer perceptron (MLP) module to finally classify the input into multiple categories.
Finally, the twelfth contribution, the last paper in the issue, presents a systematic literature review (SLR) of studies on traffic sign detection and recognition using YOLO published in the years 2016–2022, highlighting the following aspects: the most common applications are vehicular security and intelligent and autonomous vehicles; the majority of the sign datasets used to train, test, and validate YOLO-based systems are publicly available, with emphasis on datasets from Germany and China; and the most popular desktop data processing hardware are NVIDIA RTX 2080, Titan Tesla V100 and Jetson Xavier NX.
The Guest Editors would like to extend their sincere gratitude to the authors for their valuable and high-quality contributions to this Special Issue, and also to the anonymous reviewers for their insightful comments and professional evaluations. In addition, we acknowledge the excellent support provided by the MDPI associate editors in bringing this project to completion, especially the great support of the Managing Editor of this Special Issue, Ms. Kelly Su.
Conflicts of Interest
The author declares no conflict of interest.
List of Contributions
- Firan, D.O. D3SSTrack: Center-Focused State-Space Modeling for Monocular 3D Multi-Object Tracking. Mathematics 2026, 14, 1737. https://doi.org/10.3390/math14101737.
- Fatahi, M.; Zadeh, D.S.; Noormohammadi-Asl, A.; Moshiri, B.; Basir, O.A.; Sadjadi, E.N.; García, J.; Molina, J.M. CB-OWL-ViT: A Multimodal Cost-Effective Framework for Contagious Disease Monitoring. Mathematics 2026, 14, 647. https://doi.org/10.3390/math14040647.
- Mahmoud, M.; Senussi, M.F.; Abdalla, M.; Kasem, M.S.; Kang, H.S. M2UNet: A Segmentation-Guided GAN with Attention-Enhanced U2-Net for Face Unmasking. Mathematics 2026, 14, 477. https://doi.org/10.3390/math14030477.
- Kasem, M.S.; Mahmoud, M.; Senussi, M.F.; Abdalla, M.; Kang, H.S. KORIE: A Multi-Task Benchmark for Detection, OCR, and Information Extraction on Korean Retail Receipts. Mathematics 2026, 14, 187. https://doi.org/10.3390/math14010187.
- Zhu, F.L.M.; Zhao, M.; Sun, Y.; Wu, W. MSLCP-DETR: A Multi-Scale Linear Attention and Sparse Fusion Framework for Infrared Small Target Detection in Vehicle-Mounted Systems. Mathematics 2026, 14, 67. https://doi.org/10.3390/math14010067.
- Wang, Y.; Wang, Y.; Li, W.; Liang, Y.; Huang, L.; Ning, X. LV-FeatEx: Large Viewpoint-Image Feature Extraction. Mathematics 2025, 13, 1111. https://doi.org/10.3390/math13071111.
- Huang, J.H.; Chen, Y.H.; Tsai, Y.L. Utilizing Cross-Ratios for the Detection and Correction of Missing Digits in Instrument Digit Recognition. Mathematics 2024, 12, 1669. https://doi.org/10.3390/math12111669.
- Akhtar, A.; Muhammad, R.A.; Yousaf, H.; Velastin, S.A. Real-Time Motorbike Detection: AI on the Edge Perspective. Mathematics 2024, 12, 1103. https://doi.org/10.3390/math12071103.
- Dobrzycki, A.D.; Bernardos, A.M.; Bergesio, L.; Pomirski, A.; Sáez-Trigueros, D. Exploring the Use of Contrastive Language-Image Pre-Training for Human Posture Classification: Insights from Yoga Pose Analysis. Mathematics 2024, 12, 76. https://doi.org/10.3390/math12010076.
- Zhao, F.H.W.; Fan, X. Automatic Recognition of Indoor Fire and Combustible Material with Material-Auxiliasry Fire Dataset. Mathematics 2024, 12, 54. https://doi.org/10.3390/math12010054.
- Lin, X.; Huang, T.; Ruan, Z.; Yang, X.; Chen, Z.; Zheng, G.; Feng, C. Automatic Evaluation of Functional Movement Screening Based on Attention Mechanism and Score Distribution Prediction. Mathematics 2023, 11, 4936. https://doi.org/10.3390/math11244936.
- Flores-Calero, M.; Astudillo, C.A.; Guevara, D.; Maza, J.; Lita, B.S.; Defaz, B.; Ante, J.S.; Zabala-Blanco, D.; Armingol-Moreno, J.M. Traffic Sign Detection and Recognition Using YOLO Object Detection Algorithm: A Systematic Review. Mathematics 2024, 12, 297. https://doi.org/10.3390/math12020297.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.