Point cloud alignment plays a crucial role in numerous fields and applications, enabling the integration, analysis, and reconstruction of 3D data. SAmple Consensus Initial Alignment (SAC-IA), a coarse alignment algorithm, provides the initial position for precise alignment. While SAC-IA can effectively capture both global and local information, it incurs high computational costs associated with the correspondence matching of high-dimensional features. To address this challenge, we propose FSAC-IA, a faster SAC-IA-based algorithm designed to accelerate LiDAR point cloud alignment. FSAC-IA introduces a light Fast Point Feature Histogram (FPFH) descriptor to optimize the feature presentation, ensuring high matching speed without sacrificing accuracy. We also replace the original k-d tree structure in the SAC-IA, with a Hierarchical Navigable Small World (HNSW) structure for descriptor-matching, enabling efficient large-scale point cloud alignment. Experiments on fine-grained LiDAR point clouds demonstrate that FSAC-IA achieves a 7x speedup over traditional k-d tree-based alignment algorithms, establishing a new benchmark for efficient and accurate point cloud alignment.

Knowledge graphs (KGs) have been shown to be effective in improving recommendation quality by introducing rich item properties as auxiliary information. The success of current KG-based recommender systems (RSs) lies in the capability of modeling high-quality item representations. This is achieved by identifying significant properties for items and exploring the intrinsic correlation between items on the KG. However, since current KG-based works only focus on learning user implicit knowledge from KGs through items, the limited user-item interaction behavior is still an obstacle to learning high-quality user representations. Furthermore, irrelevant connections in the KG may lead to erroneous messaging during the process of high-order graph feature learning of users and items. This could subsequently result in the inaccurate recommendation of items to users. To overcome above limitations, we propose a Review-enhanced Contrastive Learning on KGs (RCLKG) model for high-quality recommendation. We first construct a review-enhanced KG by exploring user explicit preferences in reviews with the extracted review entities. Then, we design a review-aware self-augmentation mechanism that seamlessly integrates explicit review knowledge with item-aligned KGs to discard irrelevant neighbor nodes of users and items. Furthermore, we develop a global-level graph aggregation schema with a refined constraint on the merged denoising KG to further optimize the denoising KG generation by considering high-order connections with less erroneous messaging. Finally, experimental results on the rating prediction and the click-through rate prediction (CTR) tasks with three real-word datasets demonstrate the superiority of our proposed RCLKG model in comparison with the state-of-the-art baselines. • A novel review-enhanced contrastive learning model on KGs called RCLKG is proposed. • Review knowledge is injected into the knowledge graph as explicit knowledge of users. • Review-enhanced self-augmentation mechanism filters out irrelevant nodes in the KG. • Global-level graph encoder on merged denoising KG refines denoising KG generation. • Extensive experiments demonstrate the superiority of RCLKG.

The Iterative Closest Points (ICP) algorithm is the most widely used method for estimating rigid transformation in 3D point cloud registration. However, the ICP relies on repeatedly performing computationally intensive nearest neighbor searches (NNS) within 3D space. This dependency becomes a significant bottleneck when processing large datasets, thereby hindering the practical deployment of point cloud technologies in real-world applications. To address this issue, we propose two approximate nearest neighbor search (ANNS) acceleration strategies for efficient improvement of the processing speed of the NNS. Our strategies first voxelize target cloud points and then fill voxels in the 3D coordinate space around the source point cloud in two different ways, which can convert the global nearest neighbor search to a local search. Both the proposed methods are suited to be parallelized on GPUs with a low computational load. Extensive experiments show that our methods significantly accelerate NNS processing while maintaining high accuracy, outperforming most of the currently known approaches.

The 3D Non-Local Means (NLM) algorithm has become a crucial preprocessing technique for 3D image datasets due to its effectiveness in denoising while preserving fine details. This method has been proven to be highly efficient in high-demand tasks within industrial applications such as medical imaging and remote sensing. The 3D NLM algorithm computes the filtered value for each voxel by calculating the weighted average of all voxels within a 3D search window, where the weights are determined by the similarity between pairs of 3D template windows. Therefore, the computational burden becomes significant, especially in embedded GPUs with limited computational power and memory resources. To address this issue, we propose an efficient GPU parallel kernel to minimize redundant computations and memory accesses. The kernel integrates three nested reuse strategies to handle redundant computations in three dimensions: for columns, we leverage the fast data exchange mechanism to reuse column computation results via on-chip registers; for rows, we use a sliding window strategy, utilizing GPU global memory as an intermediary to store and reuse similarity values between filtered rows; and for channels, we introduce a zigzag scanning strategy that enables simultaneous computation across multiple channels and employs on-chip registers to facilitate channel computation reuse. Experimental results demonstrate that our kernel achieves an average speedup of 7.7x on the embedded Jetson AGX Xavier platform across a range of 3D image datasets compared to existing methods, showcasing exceptional performance.

The visual-based SLAM (Simultaneous Localization and Mapping) is a technology widely used in applications such as robotic navigation and virtual reality, which primarily focuses on detecting feature points from visual images to construct an unknown environmental map and simultaneously determines its own location. It usually imposes stringent requirements on hardware power consumption, processing speed, and accuracy. Currently, the ORB (Oriented FAST and Rotated BRIEF)-based SLAM systems have exhibited superior performance in terms of processing speed and robustness. However, they still fall short of meeting the demands for real-time processing on mobile platforms. This limitation is primarily due to the time-consuming Oriented FAST calculations accounting for approximately half of the entire SLAM system. This article presents two methods to accelerate the Oriented FAST feature detection on low-end embedded GPUs. These methods optimize the most time-consuming steps in Oriented FAST feature detection: FAST feature point detection and Harris corner detection, which is achieved by implementing a binary-level encoding strategy to determine candidate points quickly and a separable Harris detection strategy with efficient low-level GPU hardware-specific instructions. Extensive experiments on a Jetson TX2 embedded GPU demonstrate an average speedup of over 7.3 times compared to widely used OpenCV with GPU support. This significant improvement highlights its effectiveness and potential for real-time applications in mobile and resource-constrained environments.

We propose a high-performance$k$-way in-place merging algorithm by CPU-GPU cooperative processing. Current merging algorithms are either$k$-way in-place with CPUs only or in-place for data smaller than the GPU memory size. To solve this problem, we extend these merging algorithms to be more adaptable to a wider range of applications by improving the segmentation, reordering, and inter-segment merging methods used in the current merging algorithms by using both CPUs and GPUs cooperatively. We conducted experiments to evaluate the performance of our proposed algorithm and analyzed its detailed behavior by changing parameters and data distributions. We also applied the proposed merging algorithm to the merge sort algorithm to verify its performance.

A scientific paper recommender system (RS) is very helpful for literature searching in that it (1) helps novice researchers explore their own field and (2) helps experienced researchers explore new fields outside their area of expertise. However, existing RSs usually recommend relevant papers based on users’ static interests, i.e., papers they cited in their past publication(s) or reading histories. In this paper, we propose a novel recommendation task based on users’ dynamic interests during their paper-writing activity. This dynamism is revealed in (for example) the topic shift while writing the Introduction vs. Related Works section. In solving this task, we developed a new pipeline called “ S ection-based S equential S cientific Pa per R ecommendation (S3PaR)”, which recommends papers based on the context of the given user’s currently written paper section. Our experiments demonstrate that this unique task and our proposed pipeline outperform existing standard RS baselines.

The GIST descriptor is a classic feature descriptor primarily used for scene categorization and recognition tasks. It drives a bank of Gabor filters, which respond to edges and textures at various scales and orientations to capture the spatial structures in an image. Compared to other scene recognition algorithms that rely on detailed object detection, GIST has lower computational complexity, allowing it to be widely applied. However, its internal multi-scale and multi-orientation Gabor filters also mean that systems based on it cannot be executed fast enough. This article proposes an optimized GPU kernel for the GIST descriptor. It fully takes advantage of the symmetry of Gabor filters and proposes different optimization strategies for both oblique and orthogonal orientations. Extensive experiments demonstrate that the proposed kernel is adaptable to images of various scales and different GPUs. Compared to the cuFFT library, our kernel achieves 12.09× and 3.86× acceleration on an RTX 3080 GPU and a Jetson AGX Xavier GPU, respectively.

We propose a lightweight system for stereo-matching using embedded graphic processing units (GPUs). The proposed system overcomes the trade-off between accuracy and processing speed in stereo matching, thus further improving the matching accuracy while ensuring real-time processing. The basic idea is to construct a tiny neural network based on a variational autoencoder (VAE) to achieve the upscaling and refinement a small size of coarse disparity map. This map is initially generated using a traditional matching method. The proposed hybrid structure maintains the advantage of low computational complexity found in traditional methods. Additionally, it achieves matching accuracy with the help of a neural network. Extensive experiments on the KITTI 2015 benchmark dataset demonstrate that our tiny system exhibits high robustness in improving the accuracy of coarse disparity maps generated by different algorithms, while running in real-time on embedded GPUs.

Abstract In this paper, we propose a knowledge-aware attentional neural network (KANN) for dealing with movie recommendation tasks by extracting knowledge entities from movie reviews and capturing understandable interactions between users and movies at the knowledge level. In most recommendation systems, review information is already widely utilized to uncover the explicit preferences of users for items, especially for domains including movie recommendations, music recommendations, and book recommendations, as reviews are full of knowledge entities relevant to the domain. When processing review information, current methods usually use word embeddings to represent reviews for modeling users and items. As a result, they may split the meaning of a phrase, and thereby induce erroneous predictions. Moreover, most methods capture high-order interactions between users and items after obtaining latent low-dimensional representations, which means they cannot discover understandable interactions or provide knowledge-level explanations. By incorporating knowledge graph representation into movie recommendation tasks, the proposed KANN can not only capture the inner attention among user (movie) reviews but also compute the outer attention values between users and movies before generating corresponding latent vector representations. These characteristics enable the explicit preferences of users for movies to be learned and understood. We test our model on two datasets (IMDb and Amazon) for the movie rating prediction task and the click-through rate prediction task and show that it outperforms some of the existing state-of-the-art models and gains outstanding prediction performances in cases with a very small amount of reviews. Furthermore, we demonstrate the high explainability of the proposed KANN by visualizing the interaction between users and movies through a case study. Our results and analyses highlight the relatively high effectiveness and reliability of KANN for movie recommendation tasks.

In this paper, we propose new methods for generating playlists with a single graph, which represents multiple types of relations in a playlist. Although current users are familiar with online music services, they have difficulty in deciding which tracks to listen to because there are millions of tracks available on such services. Automated playlist generation is one of the best solutions to solving this costly task of finding interesting tracks from the enormous tracks. Accordingly, one playlist-generation task, namely, hit rate, in which several tracks are given as a user query, is focused on in this study. There are four types of context objects (playlists, tracks, artists, and users) in the basic information on playlists, and three types of relations (playlists contain tracks and artists, users create playlists and artists play and/or sing tracks) in playlists. First, different types of relations in playlists are combined, and a single graph linking different context objects is generated. Next, a random walk is applied to the graph, and the expected values of track nodes are calculated on the basis of the transition probabilities of nodes in the graph. Finally, tracks are recommended in order of the expected values. The results of an experimental evaluation of the proposed methods in comparison with conventional methods revealed that one of the proposed methods (RW-hybrid) improved effectiveness by up to 21%. Moreover, this method reduces execution time as much as the fastest existing methods.

In this study, we propose a readability measure for Web documents and an information retrieval system that considers readability. Previous information retrieval systems aim to identify documents that are relevant to a given query; however, as information requirements of search system users becomes increasingly diverse and complicated, systems that take such new criteria into account are constantly being introduced. In particular, the focus of our present paper is on readability. Given that the population of non-native English speakers exceeds that of native English speakers, incorporating readability into an information retrieval system is crucial. Therefore, we propose (1) a readability measure that considers document simplicity and document structure as new features for readability and (2) a score fusion method that combines relevance and readability scores. In our experimental results, we found that our proposed readability measure outperformed an existing readability measure. Moreover, we found score fusion methods using a statistical framework called a copula improved overall accuracy as compared to such existing methods as linear combination.

We demonstrate the geometrically-correct projection-based texture mapping onto a deformable object like a cloth. This system can be used to simulate design that involves change in shape, such as sheets of malleable material. The geometrically-correct projection-based texture mapping onto a cloth is conducted using the measurement of object's 3D shape and the detection of the retro-reflective marker on the object's surface. Rapid prototyping is used as an example application of this projection technique.

In this paper, we propose a method for estimating the camera pose for an environment in which the intrinsic camera parameters change dynamically. In video see-through augmented reality (AR) technology, image-based methods for estimating the camera pose are used to superimpose virtual objects onto the real environment. In general, video see-through-based AR cannot change the image magnification that results from a change in the camera׳s field-of-view because of the difficulty of dealing with changes in the intrinsic camera parameters. To remove this limitation, we propose a novel method for simultaneously estimating the intrinsic and extrinsic camera parameters based on an energy minimization framework. Our method is composed of both online and offline stages. An intrinsic camera parameter change depending on the zoom values is calibrated in the offline stage. Intrinsic and extrinsic camera parameters are then estimated based on the energy minimization framework in the online stage. In our method, two energy terms are added to the conventional marker-based method to estimate the camera parameters: reprojection errors based on the epipolar constraint and the constraint of the continuity of zoom values. By using a novel energy function, our method can accurately estimate intrinsic and extrinsic camera parameters. We confirmed experimentally that the proposed method can achieve accurate camera parameter estimation during camera zooming.

Projection-based Augmented Reality commonly employs a rigid substrate as the projection surface and does not support scenarios where the substrate can be reshaped. This investigation presents a projection-based AR system that supports deformable substrates that can be bent, twisted or folded. We demonstrate a new invisible marker embedded into a deformable substrate and an algorithm that identifies deformations to project geometrically correct textures onto the deformable object. The geometrically correct projection-based texture mapping onto a deformable marker is conducted using the measurement of the 3D shape through the detection of the retro-reflective marker on the surface. In order to achieve accurate texture mapping, we propose a marker pattern that can be partially recognized and can be registered to an object’s surface. The outcome of this work addresses a fundamental vision recognition challenge that allows the underlying material to change shape and be recognized by the system. Our evaluation demonstrated the system achieved geometrically correct projection under extreme deformation conditions. We envisage the techniques presented are useful for domains including prototype development, design, entertainment and information based AR systems.

Augmented reality (AR) x-ray interaction is an enabling technology for providing students with virtual abstractions of the interior of an object. It provides students contextual visualization - the presentation of virtual information in the rich context of a real environment - thereby offering compelling experiences. According to experiential learning theory, such personal experiences are necessary for reaching different types of learners. AR x-ray is a novel interaction technique for education, thus, it is necessary to investigate how it affects the students' perception. We implemented AR x-ray using a state-of-the-art occlusion technique, and compared it to viewing 3D objects without occlusion. Results of two user studies (n=23 and n=47) show that there are no significant differences in realism, perception of depth, and visibility with occlusion and without occlusion, and that the current technique is usable for educational purposes. We also conducted interviews with both students (n=23) and teachers (n=12). Results indicate that AR x-ray is perceived to be useful for motivating and explaining to students. The teachers expressed willingness to adopt AR x-ray and to undergo training for using AR-based teaching materials.

Engineers and educators alike have prototyped a variety of augmented reality learning experiences (ARLEs). However, adapting ARLEs in educational practice would require an interdisciplinary approach that considers learning theory, pedagogy and instructional design. To address this requirement, we model ARLEs as learning objects by outlining the necessary components, and we propose a participatory design to demonstrate the authoring process of an augmented reality learning object (ARLO). ARLOs can be made useful in many scenarios if teachers are empowered to edit its context elements, content and instructional activity. Lastly, we point to the research questions entailed in modeling ARLEs as ARLOs.

In general, video see-through based augmented reality (AR) cannot change the magnification of camera zooming parameter due to the difficulty of dealing with changes in intrinsic camera parameters. To realize the usage of camera zooming in AR, we propose a novel simultaneous intrinsic and extrinsic camera parameter estimation method based on an energy minimization framework. Our method is composed of the online and offline stages. An intrinsic camera parameter change depending on the zoom values is calibrated in the offline stage. Intrinsic and extrinsic camera parameters are then estimated based on the energy minimization framework in the online stage. In our method, two energy terms are added to the conventional marker-based camera parameter estimation method. One is reprojection errors based on the epipolar constraint. The other is the constraint of continuity of zoom values. By using a novel energy function, our method can estimate accurate intrinsic and extrinsic camera parameters. In an experiment, we confirmed that the proposed method can achieve accurate camera parameter estimation during camera zooming.

A method for fast and incremental indexing, with both effective and efficient query processing, is proposed for XML element retrieval. When frequent document updates occur on the Web, they must be handled to maintain the effectiveness of the search system. When new topics are added and document statistics change drastically, search accuracy is also reduced. We therefore consider a method not only for updating indices efficiently but also for processing queries effectively and efficiently. We construct indices for fast updating and propose a method for computing accurate term weights even under dynamically changing statistics. Experimental results show that our proposed system can handle document updates at low cost and search documents accurately even when their statistics change.

The objective of this study is to investigate the relationship between the features of augmented reality (AR) and human memorization ability. The basis of this relation is derived from the following features. The AR feature is that AR can provide information associated with specific locations in the real world. The feature of human memory is that humans can easily memorize information if the information is visually associated with specific locations. To investigate this relation, we conduct a pilot user study in which blocks are picked from some drawers. As a result, significant differences are found between a situation in which visual information is displayed at the location of each drawer in the real world and that in which textual information is displayed at an unrelated location.

This paper presents an application of Augmented Reality (AR) in Rapid Prototyping (RP) of non-textured rigid curved surfaces. By enhancing the prototypes with AR, evaluation of its design and aesthetic concepts in real-time becomes easier, saving time and production costs. In our application, no fiducial markers are required and the CAD model used to build the prototype is applied in an edge-based tracking system specially designed to deal with curved shapes. Results from a pilot user study comparing the use of a 3D software and the proposed application are also presented.

A photomosaic is a type of decorative art made up from various other photographs. We present a method for quickly generating photomosaics and propose an interactive recursive photomosaic system. Users can operate the system by using a large display with a touch input function, which allows them to alter the appearance of the image dynamically.

In this paper, we propose a non-blocking buffer management scheme based on a lock-free variant of the GCLOCK page replacement algorithm. Concurrent access to the buffer management module is a major factor that prevents database scalability to processors. Therefore, we propose a non-blocking scheme for bufferfix operations that fix buffer frames for requested pages without locks by combining Nb-GCLOCK and a non-blocking hash table. Our experimental results revealed that our scheme can obtain nearly linear scalability to processors up to 64 processors, although the existing locking-based schemes do not scale beyond 16 processors.

We propose and evaluate a method for obtaining more accurate search results in extensible markup language (XML) fragment search, which is a search that produces only relevant fragments or portions of an XML document. The existing approaches generate a ranked list in descending order of each XML fragment's relevance to a search query; however, these approaches often extract irrelevant XML fragments and overlook more relevant fragments. To address these problems, our approach extracts relevant XML fragments by considering the size of the fragments and the relationships between the fragments. Next, we score the XML fragments to generate a refined ranked list. For scoring, we rank the XML fragments that are informative for user information needs as high in the list. In particular, each XML fragment is scored using the statistics of its descendant and ancestor XML fragments.

We present a design and implementation of tracing document history and its application to detecting abnormal document access patterns. We have been working on document history system, which enables us to trace complete access history of operations for both electronic and paper documents. Document history system organizes access pattern trees based on dependency of user operations on documents. We also present a new probabilistic model, multi-layered tree Markov model (MTMM), in order to analyze access pattern trees of documents. The main feature of MTMM is to conduct tree pattern summarization and multilevel structural abstraction for trees. Our contribution is multi-level structural abstraction by incorporating hierarchical inner-state structure of Markov model. Abnormal patterns, such as sending documents to unexpected users, can be detected by applying MTMM to access pattern trees of document history system. We clarify the feasibility and effectiveness of our MTMM by several experiments for synthetic trees and real document access pattern trees.

We propose an efficient distributed query processing method for large XML data by partitioning and distributing XML data to multiple computation nodes. There are several steps involved in this method; however, we focused particularly on XML data partitioning and dynamic relocation of partitioned XML data in our research. Since the efficiency of query processing depends on both XML data size and its structure, these factors should be considered when XML data is partitioned. Each partitioned XML data is distributed to computation nodes so that the CPU load can be balanced. In addition, it is important to take account of the query workload among each of the computation nodes because it is closely related to the query processing cost in distributed environments. In case of load skew among computation nodes, partitioned XML data should be relocated to balance the CPU load. Thus, we implemented an algorithm for relocating partitioned XML data based on the CPU load of query processing. From our experiments, we found that there is a performance advantage in our approach for executing distributed query processing of large XML data.

The purpose of this study is to propose Context-Aware Support Vector Machine (C-SVM) for application in a context-dependent recommendation system. It is important to consider users’ contexts in information recommendation as users’ preference change with context. However, currently there are few methods which take into account users’ contexts (e.g. time, place, the situation and so on). Thus, we extend the functionality of a Support Vector Machines (SVM), a popular classifier method used between two classes, by adding axes of context to the feature space in order to consider the users’ context. We then applied the Context-Aware SVM (C-SVM) and the Collaborative Filtering System with Context-Aware SVM (C-SVM-CF) to a recommendation system for restaurants and then examined the effectiveness of each approach.

Summary form only given. Starting from the history of document media, especially from the point of the relationship between document media and human bodies, document media can be seen as extension of body. We see how document media have been evolved with human interaction and how human bodies have been extended by document media. But recently, we have lost those historical obtained senses in electronic documents. Some thoughts for recovering those senses are presented. Next, an example of document media in workplace is described. We see how workplace affects the interaction between document media and human being as services. Finally, the future of document media and its services are briefly forecasted.

We present a new approach to segment and visualize informally captured multi-stream meeting videos. We process the visual content in each stream individually by analyzing the differences between frames in each sequence to find change areas. These results are combined with face detection to determine visual activity in each of the streams. We then combine the activity scores from multiple streams and automatically generate a 3D representation of the video. Our representation allows the user to obtain an at-a-glance view of the video at different granularities of activity, view multiple streams simultaneously, and select particular points in time for viewing. We present experiments that suggest that low-level visual analysis can be effective for finding highlights that can be used for browsing multi-stream meeting videos.

The Fat-Btree is expected to be used as a parallel directory structure for high performance and highly reliable data intensive systems, such as databases and file systems. Though the Fat-Btree is essentially high performance, it needs to be highly reliable as well. In order to achieve high reliability, we introduce five configuration methods. We show not only their steady-state and computational availabilities but also cost related comparisons, so as to evaluate the properties of these configurations.

We present CandyTop Interface, a visualization method with positive attention for growing multimedia documents. Our method is a visualization method for growing documents by allocating changes of multimedia documents according to time line in three-dimensional space. The main feature of our method is to enable provision of positive attention on key changes of multimedia documents by visualizing multimedia documents with priority of changes for multimedia documents. By our method, we can realize multimedia systems providing at-a-glance view for multiple versions of multimedia documents. We discuss the feasibility of our method by applications to visualization for growing documents with collaborative editing in virtual spaces and for multiview video streams of technical meeting.

We propose a parallel directory structure, Fat-Btree, to improve high speed access for parallel database systems in shared nothing environments. The Fat-Btree has a threefold aim: to provide an indexing mechanism for fast retrieval in each processor; to balance the amount of data among distributed disks, and to reduce synchronization costs between processors during update operations. We use a probability based model to compare the throughput and response time of the Fat-Btree with two ordinary parallel Btree structures, with copies of a whole Btree in each processor and storing index nodes in a processor. The comparison results indicate that the Fat-Btree is suitable for actual parallel database systems that accept update operations.