Türkiye tarih boyunca pek çok farklı medeniyete ev sahipliği yapmış bir ülke olması dolayısıyla bu medeniyetlere ait, oldukça çeşitli kültürel miras ögelerini içinde barındırmaktadır. Bu bağlamda, sahip olduğumuz bu kültürel mirasın korunması ve gelecek nesillere ulaştırılması oldukça önem arz etmektedir. Kültürel miras ögeleri zaman içerisinde birçok farklı sebepten ötürü deformasyona uğrayabilmekte veya tamamen yok olabilmektedir. Bu durumun önüne geçilebilmesi için öncelikle bu kültürel miras ögelerinin 3 Boyutlu (3B) dokümantasyonunun (belgelenmesinin) yapılması gerekmektedir. Dokümantasyon sonucunda elde edilen veriler, kültürel miras ögeleri üzerindeki deformasyonların izlenmesi, restorasyon çalışmaları gibi pek çok farklı alanda katkı sunabilmektedir. Günümüzde dokümantasyon çalışmalarında kullanılan farklı yöntemler mevcuttur. Arazi üzerinde veri toplaması aşamasında kullanılan yöntemlerden birkaçı; yersel fotogrametri, hava fotogrametrisi ve yersel lazer tarama (YLT) olarak gösterilebilir. Öte yandan, arazi çalışması aşamasında toplanılan verilerin işlenebileceği birçok yazılım mevcuttur. Bu çalışmada, İstanbul Fatih ilçesi sınırlarındaki Sultanahmet Meydanı’nda bulunan tarihi Dikilitaş’ın YLT tekniği kullanılarak taraması yapılmış ve üç farklı yazılım aracılığıyla nokta bulutu üretilmiştir. Bu yazılımların kültürel mirasın dokümantasyonu çalışmalarında kullanılabilirlikleri karşılaştırmalı olarak irdelenmiştir.

Pose estimation and localization within the gastrointestinal tract, particularly the small bowel, are crucial for invasive medical procedures. However, the task is challenging due to the complex anatomy, homogeneous textures, and limited distinguishable features. This study proposes a hybrid deep learning (DL) method combining Convolutional Neural Network (CNN)-based pose estimation and optical flow to address these challenges in a simulated small bowel environment. Initial pose estimation was used to assess the performance of simultaneous localization and mapping (SLAM) in such complex settings, using a custom endoscope prototype with a laser, micromotor, and miniaturized camera. The results showed limited feature detection and unreliable matches due to repetitive textures. To improve this issue, a hybrid CNN-based approach enhanced with Farneback optical flow was applied. Using consecutive images, three models were compared: Hybrid ResNet-50 with Farneback optical flow, ResNet-50, and NASNetLarge pretrained on ImageNet. The analysis showed that the hybrid model outperformed both ResNet-50 (0.39 cm) and NASNetLarge (1.46 cm), achieving the lowest RMSE of 0.03 cm, with feature-based SLAM failing to provide reliable results. The hybrid model also gained a competitive inference speed of 241.84 ms per frame, outperforming ResNet-50 (316.57 ms) and NASNetLarge (529.66 ms). To assess the impact of the optical flow choice, Lucas–Kanade was also implemented within the same framework and compared with the Farneback-based results. These results demonstrate that combining optical flow with ResNet-50 enhances pose estimation accuracy and stability, especially in textureless environments where traditional methods struggle. The proposed method offers a robust, real-time alternative to SLAM, with potential applications in clinical capsule endoscopy. The results are positioned as a proof-of-concept that highlights the feasibility and clinical potential of the proposed framework. Future work will extend the framework to real patient data and optimize for real-time hardware.

The emergence of foundation models has driven major advancements in computer vision and natural language processing, primarily due to their strong zero-shot and few-shot capabilities powered by large-scale, diverse datasets. While earlier approaches used supervised datasets, their limited scene diversity did not perform well in unseen environments. To overcome these limitations, recent works have leveraged unlabeled monocular images, which can be automatically labeled using pre-trained models. One model can be shown as Depth Anything, which demonstrated robust zero-shot performance across diverse scenarios, with Depth Anything V2 further improving accuracy. In this study, the performance of Depth Anything V1 and V2 models was evaluated in satellite-derived bathymetry using Sentinel 2 satellite imagery. The accuracy of these predicted depth maps was evaluated by comparing them with bathymetric data obtained from the National Oceanic and Atmospheric Administration’s (NOAA) National Centers for Environmental Information (NCEI) as the ground truth. The results show that the correlation between Depth Anything V1 predictions and NOAA NCEI data was 56.69%, while the correlation for Depth Anything V2 reached 84.54%. The predicted depth maps were also scaled to obtain Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). The RMSE and MAE values for Depth Anything V1 are 0.4135 m and 0.34 m, respectively, while the RMSE and MAE values for V2 are 0.2681 m and 0.2089 m, respectively. This improvement shows the capability of Depth Anything V2 in estimating underwater terrain from monocular satellite imagery, which also demonstrates its potential for cost-effective bathymetric mapping in remote sensing applications. In addition to deep learning-based approaches applied in the test area, a satellite-derived depth map was also generated using the classical band ratio method. Compared with reference bathymetric data, the correlation coefficient, RMSE, and MAE were found to be 38.20%, 0.4639m, and 0.3746m, respectively.

The long, tortuous, and tissue-homogeneous structure of the small bowel makes image-based three-dimensional (3D) modeling studies technically complex. Since wireless capsule endoscopy (WCE) systems generally produce only two-dimensional images, obtaining precise 3D structures with this limited data is highly challenging with existing traditional methods. In this study, we propose a novel capsule endoscopy prototype that enables 3D reconstruction of the small bowel from consecutive monocular images. Laser points that are circumferentially scanned and thus projected onto the intestinal surface are segmented using HSV color space transformation and area-based mask generation. The detected point coordinates are then sequentially placed according to the image order to generate a dense 3D point cloud. Unlike traditional methods, the proposed approach reconstructs 3D structure using only monocular images, without relying on stereo vision or multi-view algorithms. This point cloud can be used for visual analysis and modeling studies by reflecting the basic geometric properties of the intestinal surface. The results show that the proposed system achieves geometrically consistent 3D bowel model reconstruction with a total root mean square error (RMSE) of approximately ± 0.3 cm. First, a 3D point cloud was generated in the Python environment using sequential laser point coordinates. The point cloud generated in the Python environment achieved an RMSE of 2.27 mm. The same point cloud was then imported into CloudCompare software for independent validation. An RMSE of 3.13 mm was calculated in a standard 3D analysis tool. These results demonstrate that the proposed method achieves geometrically consistent and quantitistically reliable reconstruction under controlled experimental conditions. The proposed framework is evaluated as a proof-of-concept using controlled inter-frame depth displacement assumptions. The 3D modeling study was also conducted using Agisoft Metashape software. Additionally, the SIFT-SfM (Scale-Invariant Feature Transform-Structure-from-Motion) and ORB-SfM (Oriented FAST and Rotated BRIEF Structure-from-Motion) methods were applied in the Python environment. However, both approaches failed to produce a consistent model on the phantom data. These results demonstrate the suitability of the proposed laser-assisted monocular capsule endoscopy approach in challenging anatomical environments where visual features are limited. Additionally, we conducted a soft-tissue phantom experiment to evaluate the method under more anatomically realistic surface conditions, where the system successfully captured the general fold patterns of the tissue. The presented framework provides an initial experimental basis for future deep learning-based surface reconstruction and clinical validation studies. The processing pipeline achieves real-time–capable throughput under offline evaluation conditions. Furthermore, by providing a practical solution with low hardware requirements, it represents an important first step toward the development of more advanced and clinically applicable 3D modeling systems.

The correct documentation of the structure of cultural heritage is necessary for its conservation, restoration, and 3D analysis. This paper explores the use of professional Terrestrial Laser Scanning (TLS) and iPad Pro LiDAR sensors to create accurate three-dimensional models for documenting cultural heritage sites. The case study was conducted on the Molla Hüsrev Mosque in Istanbul, which dates back to the 15th century and is an Ottoman structure. The TLS scans were performed using the Leica P40 scanner, while the iPad Pro LiDAR scans were conducted using the SiteScape app. The two datasets were registered using the Iterative Closest Point (ICP) algorithm, which achieved an accuracy of 0.007 m. The ICP algorithm was also used to determine registration accuracy in complex buildings, and recommendations for improving accuracy in such buildings are discussed in the paper. The results showed that the TLS system achieved high geometric accuracy and high resolution (0.029 m RMS). The iPad Pro LiDAR system improved the completeness of surface reconstruction on complex surfaces. The fusion of the two datasets produced a denser, more complete dataset of the mosque. The paper's results show that while the iPad Pro LiDAR system is less accurate than the TLS system, its portability and high frame rate make it an ideal supporting tool for multi-sensor approaches. The fusion of the two datasets saves time and operational costs in the field while preserving spatial consistency. The paper demonstrates the growing use of consumer-grade LiDAR sensors in digital heritage documentation and presents an approach that scales up support for both consumer and professional sensors.

Localization can be defined as the process of determining the position and orientation of an entity within an environment, that would enable it to navigate and carry out tasks effectively. It is of fundamental importance for a wide range of areas such as robotics, medicine, indoor and outdoor navigation, autonomous vehicles, etc. Localization problems might be solved either with hardware or software designs. However, considering the challenging environments that would need a localization process, hardware designs might be complicated to apply to this problem. Indoor navigation can be shown as an example of these challenging environments since classical positioning methods cannot be used in such places. In this case, visual localization might be a solution since it requires either monocular or stereo images taken through the path. It would be a time-saving and cost-effective way to determine the locations that images were taken, thus the path that a robot or medical instrument, etc. take along the way. In this case, it is important the determine the performances of different localization algorithms. In this study, the performance of two different localization algorithms in indoor navigation was tested. In this context, Visual odometry and EKF SLAM algorithms were used to determine the camera trajectory utilizing the images that were taken in a straight corridor with a smartphone camera. To determine the accuracy of each method, the distances between each image-taking point were measured and compared with the distances obtained from the algorithm. Thus, root mean square error values were determined by each method. The precisions of each method were also given based on the fact that the distance between each image-taking point was equal. Therefore, the usage of both algorithms in indoor navigation was discussed.