Transformer-based methods have become dominant in the medical image research field since the Vision Transformer achieved superior performance. Although transformer-based approaches have resolved long-range dependency problems inherent in Convolutional Neural Network (CNN) methods, they struggle to capture local detail information. Recent research focuses on the robust combination of local detail and semantic information. To address this problem, we propose a novel transformer-CNN hybrid network named RAPUNet. The proposed approach employs MetaFormer as the transformer backbone and introduces a custom convolutional block, RAPU (Residual and Atrous Convolution in Parallel Unit), to enhance local features and alleviate the combination problem of local and global features. We evaluate the segmentation performance of RAPUNet on popular benchmarking datasets for polyp segmentation, including Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, EndoScene-CVC300, and ETIS-LaribPolypDB. Experimental results show that our model achieves competitive performance in terms of mean Dice and mean IoU. Particularly, RAPUNet outperforms state-of-the-art methods on the CVC-ClinicDB dataset. Code available: https://github.com/hyunnamlee/RAPUNet .
This paper presents a robust, efficient lane-marking feature extraction method using a graph model-based approach. To extract the features, the proposed hat filter with adaptive sizes is first applied to each row of an input image and local maximum values are extracted from the filter response. The features with the maximum values are fed as nodes to a connected graph structure, and the edges of the graph are constructed using the proposed neighbor searching method. Nodes related to lane-markings are then selected by finding a connected subgraph in the graph. The selected nodes are fitted to line segments as the proposed features of lane-markings. The experimental results show that the proposed method not only yields at least 2.2% better performance compared to the existing methods on the KIST dataset, which includes various types of sensing noise caused by environmental changes, but also improves at least 1.4% better than the previous methods on the Caltech dataset which has been widely used for the comparison of lane marking detection. Furthermore, the proposed lane marking detection runs with an average of 3.3 ms, which is fast enough for real-time applications.
Deep learning-based computer vision algorithms, especially image segmentation, have been successfully applied to pixel-level crack detection. The prediction accuracy relies heavily on detecting the performance of fine-grained cracks and removing crack-like noise. We propose a fast encoder-decoder network with scaling attention. We focus on a low-level feature map by minimizing encoder-decoder pairs and adopting an Atrous Spatial Pyramid Pooling (ASPP) layer to improve the detection accuracy of tiny cracks. Another challenge is the reduction in crack-like noise. This introduces a novel scaling attention, , to suppress irrelevant regions. However, removing crack-like noise, such as grooving, is difficult by using only improved segmentation networks. In this study, a crack dataset is generated. It contains 11,226 sets of images and masks, which are effective for detecting detailed tiny cracks and removing non-semantic objects. Our model is evaluated on the generated dataset and compared with state-of-the-art segmentation networks. We use the mean Dice coefficient (mDice) and mean Intersection over union (mIoU) to compare the performance and FLOPs for computational complexity. The experimental results show that our model improves the detection accuracy of fine-grained cracks and reduces the computational cost dramatically. The mDice score of the proposed model is close to the best score, with only a 1.2% difference but two times fewer FLOPs.
This paper presents a study on aircraft engine identification using real-time 3D LiDAR point cloud segmentation technology, a key element for the development of automated docking systems in airport boarding facilities, known as jetbridges. To achieve this, 3D LiDAR sensors utilizing a spinning method were employed to gather surrounding environmental 3D point cloud data. The raw 3D environmental data were then filtered using the 3D RANSAC technique, excluding ground data and irrelevant apron areas. Segmentation was subsequently conducted based on the filtered data, focusing on aircraft sections. For the segmented aircraft engine parts, the centroid of the grouped data was computed to determine the 3D position of the aircraft engine. Additionally, PointNet was applied to identify aircraft engines from the segmented data. Dynamic tests were conducted in various weather and environmental conditions, evaluating the detection performance across different jetbridge movement speeds and object-to-object distances. The study achieved a mean intersection over union (mIoU) of 81.25% in detecting aircraft engines, despite experiencing challenging conditions such as low-frequency vibrations and changes in the field of view during jetbridge maneuvers. This research provides a strong foundation for enhancing the robustness of jetbridge autonomous docking systems by reducing the sensor noise and distortion in real-time applications. Our future research will focus on optimizing sensor configurations, especially in environments where sea fog, snow, and rain are frequent, by combining RGB image data with 3D LiDAR information. The ultimate goal is to further improve the system’s reliability and efficiency, not only in jetbridge operations but also in broader autonomous vehicle and robotics applications, where precision and reliability are critical. The methodologies and findings of this study hold the potential to significantly advance the development of autonomous technologies across various industrial sectors.
Recent advancements in online learning platforms have improved accessibility; however, visually impaired learners still face barriers accessing visual materials such as charts and diagrams in lecture videos. When instructors provide limited or no descriptions, learners miss out on critical content. This study proposes an AI-powered web-based system that detects visual elements in lecture videos and generates interactive natural language descriptions. The system integrates a YOLO-based chart detector with a T5-based caption generator in a novel detection-captioning pipeline. Leveraging cloud deployment, it achieves up to 27x faster processing than conventional methods, ensuring smooth operation on low-spec devices. Lightweight models further enhance computational efficiency and responsiveness. Upon detecting a chart, the system pauses playback, delivers a synthesized spoken and textual description, then resumes the lecture, offering seamless access to visual content. A user study with 20 visually impaired participants guided the design of an optimized interaction flow that supports accessibility without disrupting learning engagement. Users can also review content and request additional explanations. The platform complies with WAI-ARIA guidelines, supporting screen readers, keyboard navigation, and interactive feedback for diverse users. This research contributes to accessible education by combining AI-driven visual interpretation with interactive HCI design, transforming visual content into a more inclusive, interactive learning experience.