For highly immersive mobile volumetric video streaming, it is essential to deliver photo-realistic full-scene content with smooth playback. Unlike traditional representations such as point clouds, 3D Gaussian Splatting (3DGS) has gained attention for its ability to represent high-quality full-scene 3D content. However, our preliminary experiments show that existing methods for 3DGS-based videos fail to achieve smooth playback on mobile devices. In this paper, we propose Vega, a 3DGS-based photo-realistic full-scene volumetric video streaming system that ensures real-time playback on mobile devices. The core idea behind Vega's real-time rendering is object-level selective computation, which allocates computational resources to visually important objects to meet strict rendering deadlines. To enable mobile streaming based on the selective computation, Vega addresses two challenges: (1) designing an encoding scheme that optimizes the data size of videos while being compatible with object-level prioritization, and (2) developing a rendering pipeline that efficiently operates on resource-constrained mobile devices. We implemented an end-to-end Vega system, consisting of a streaming server and an Android application. Experimental results on commodity smartphones show that Vega achieves 30 frames per second (FPS) for full-scene volumetric video streaming while maintaining competitive data size and visual quality compared to existing baselines.

Mobile Augmented Reality (AR) applications demand high-quality, real-time visual prediction, including pixel-level depth and semantics, to enable immersive and context-aware user experiences. Recently, Vision Foundation Models (VFMs) offer strong generalization capabilities on diverse and unseen data, supporting scalable mobile AR experiences. However, deploying VFMs on mobile devices is challenging due to computational limitations, particularly in maintaining both prediction accuracy and real-time performance. In this paper, we present ARIA, the first system that enables on-device inference acceleration of a VFM. ARIA employs the heterogeneity of mobile processors through a parallel and selective inference scheme: full-frame prediction is periodically offloaded to a processor with high parallelism capability like GPU, while low-latency updates on dynamic regions are conducted via a specialized accelerator like NPU. Implemented and evaluated using mobile devices, ARIA achieved significant improvements in accuracy and deadline success rate on diverse real-world mobile AR scenarios.

Several studies have analyzed traffic patterns using Vehicle Detector (VD) and Global Positioning System (GPS) data. VD records the speed of vehicles passing through detectors, GPS data captures traffic speed on the roads. However, unobserved data gaps may arise due to physical malfunctions of sensors in VD data or interruptions in satellite signal reception for GPS data. Unobserved data adds complexity to the analysis and prediction of urban traffic networks. To tackle this challenge, researchers have attempted to estimate unobserved data using spatiotemporal patterns, but approaches that rely solely on past time points are inherently less reliable. In this study, we propose a Virtual network-based Diffusion Convolutional Recurrent Neural Network (V-DCRNN) for estimating unobserved speed data in urban traffic networks using a virtual network. The virtual network is created by adding nodes and edges in virtual directions based on observed nodes at intersections, thereby enhancing the traffic network. The V-DCRNN, which utilizes the diffusion convolution process in the virtual network, uses the augmented traffic network as input to predict traffic speed. Unobserved speed data is estimated based on the values of virtual nodes predicted by V-DCRNN. We evaluate the proposed V-DCRNN model through unobserved speed data estimation experiments in the urban traffic network, where we randomly mask individual nodes and entire intersections. The main contributions of this work are as follows: 1) the design of a virtual network to model additional traffic dynamics at intersections, facilitating the estimation of unobserved speed data; 2) the development of the V-DCRNN model, which leverages a self-attention mechanism to capture spatiotemporal dependencies in urban traffic networks by incorporating the virtual network as input; and 3) an evaluation of the V-DCRNN’s ability to estimate unobserved speed data in urban traffic networks with up to 20% unobserved nodes, demonstrating robust and reliable performance.

The transportation network is an important element in an urban system that supports daily activities, enabling people to travel from one place to another. One of the key challenges is the network complexity, which is composed of many node pairs distributed over the area. This spatial characteristic results in the high dimensional network problem in understanding the 'cause' of problems such as traffic congestion. Recent studies have proposed visual analytics systems aimed at understanding these underlying causes. Despite these efforts, the analysis of such causes is limited to identified patterns. However, given the intricate distribution of roads and their mutual influence, new patterns continuously emerge across all roads within urban transportation. At this stage, a well-defined visual analytics system can be a good solution for transportation practitioners. In this paper, we propose a system, CATOM (Causal Topology Map), for the cause-effect analysis of traffic patterns based on Granger causality for extracting causal topology maps. CATOM discovers causal relationships between roads through the Granger causality test and quantifies these relationships through the causal density. During the design process, the system was developed to fully utilize spatial information with visualization techniques to overcome the previous problems in the literature. We also evaluate the usability of our approach by conducting a SUS(System Usability Scale) test and traffic cause analysis with the real-world data from two study sites in collaboration with domain experts.

Researchers have been actively engaged in recent studies utilizing the denoising diffusion probabilistic model (DDPM), a generative model that shows stable convergence during training, unlike other generative models, and represents data distributions that approximate real-world data. Due to these advantages, DDPM is widely employed for image, video, and audio generation. Recent DDPM research extends beyond image, video, and audio to various datasets and applications. The majority of real-world scenarios involve spatiotemporal data. Consequently, researchers propose prediction models using spatiotemporal data across domains such as traffic forecasting, weather prediction, and disease tracking. While most spatiotemporal data prediction models focus on traffic data, weather and epidemic data present higher uncertainty due to regional variations, seasonal changes, fluctuations in infection rates, and uncertainties in reported cases. To address these challenges of spatiotemporal data, we propose a DDPM-based model. In contrast to previous DDPM studies, we propose a denoising network combining Graph Structure Learning (GSL) to utilize the dynamic nature of spatiotemporal data. We assess performance using three real-world datasets encompassing traffic, weather, and epidemics and conduct an ablation study comparing GSL methods.

Researchers have been persistently working to address the issue of missing values in time series data. Numerous models have been proposed, striving to estimate the distribution of the data. The Radial Basis Functions Neural Network (RBFNN) has recently exhibited exceptional performance in estimating data distribution. In this paper, we propose a time series imputation model based on RBFNN. Our imputation model learns local information from timestamps to create a continuous function. Additionally, we incorporate time gaps to facilitate learning information considering the missing terms of missing values. We name this model the Missing Imputation Multivariate RBFNN (MIM-RBFNN). However, MIM-RBFNN relies on a local information-based learning approach, which presents difficulties in utilizing temporal information. Therefore, we propose an extension called the Missing Value Imputation Recurrent Neural Network with Continuous Function (MIRNN-CF) using the continuous function generated by MIM-RBFNN. We evaluate the performance using two real-world datasets with non-random missing and random missing patterns, and conduct an ablation study comparing MIM-RBFNN and MIRNN-CF.

3D object detection with omnidirectional views enables safety-critical applications such as mobile robot navigation. Such applications increasingly operate on resource-constrained edge devices, facilitating reliable processing without privacy concerns or network delays. To enable cost-effective deployment, cameras have been widely adopted as a low-cost alternative to LiDAR sensors. However, the compute-intensive workload to achieve high performance of camera-based solutions remains challenging due to the computational limitations of edge devices. In this paper, we present Panopticus, a carefully designed system for omnidirectional and camera-based 3D detection on edge devices. Panopticus employs an adaptive multi-branch detection scheme that accounts for spatial complexities. To optimize the accuracy within latency limits, Panopticus dynamically adjusts the model's architecture and operations based on available edge resources and spatial characteristics. We implemented Panopticus on three edge devices and conducted experiments across real-world environments based on the public self-driving dataset and our mobile 360° camera dataset. Experiment results showed that Panopticus improves accuracy by 62% on average given the strict latency objective of 33ms. Also, Panopticus achieves a 2.1× latency reduction on average compared to baselines.

In this paper, we present a LiDAR Inertial Odometry (LIO) algorithm utilizing adaptive keyframe generation which achieves fast and accurate state estimation for aerial and ground robots. It is known that keyframe generation significantly affects the performance of Simultaneous Localization and Mapping (SLAM) algorithms. Unlike existing SLAM algorithms that generate keyframes based on fixed conditions, we propose to use adaptive keyframe generation conditions considering characteristics of surrounding environment using real-time LiDAR scans. When a keyframe is generated, the keyframe and the corresponding LiDAR measurements are stored in our novel data structure designed for efficient sub- map generation. The scan to sub-map matching module then uses the Generalized Iterative Closest Point (GICP) algorithm to adjust estimated states at a global scale, producing more accurate and globally consistent state estimation results even in large-scale underground environments. Experimental results from diverse types of underground environments show that the proposed method outperforms the existing state-of-the-art LIO algorithms in various metrics such as computational speed, CPU usage, and accuracy.

Autonomous driving in an urban environment is one of the most actively studied topics. To date, many studies on autonomous driving can be classified into two main approaches: 1.local perception-based approach 2.global path tracking-based approach. However, each approach has its own limitations for fully autonomous driving. In the case of perception-based approach, it is impossible to autonomously drive to the global destination because it only runs locally within the sensor range. On the other hand, the path tracking-based approach relies heavily on accurate navigation. For accurate navigation, there are many studies using expensive equipment or extremely precise and detailed maps, but they have not been resolved yet, and they are also not practical. In this paper, we address the problem of autonomous driving in an urban environment through the proposed hybrid control architecture. Proposed control architecture is designed to be complementary of local perception-based and global path tracking-based approaches. Especially, end-to-end deep learning based autonomous driving, which mimics human driving, is applied as a local perceptionbased approach. In addition, path tracking-based autonomous driving is performed in an environment where directional information to the destination is required, such as intersections. At the same time, our hybrid control architecture effectively compensates for the navigation error using ICP matching between the perception-based driven trajectory and the global path to the destination without any highly detailed prior map or expensive equipment. The performance of the proposed architecture on a full-scale autonomous vehicle is verified through experiments in the urban environment.