ORBBuf: A Robust Buffering Method for Remote Visual SLAM

Simultaneous localisation and mapping (SLAM) is crucial in mobile robotics. Most visual SLAM systems assume that the environment is static. However, in real life, there are many dynamic objects, which affect the accuracy and robustness of these systems. To improve the performance of visual SLAM systems, this study proposes a dynamic visual SLAM (SEG-SLAM) system based on the orientated FAST and rotated BRIEF (ORB)-SLAM3 framework and you only look once (YOLO)v5 deep-learning method. First, based on the ORB-SLAM3 framework, the YOLOv5 deep-learning method is used to construct a fusion module for target detection and semantic segmentation. This module can effectively identify and extract prior information for obviously and potentially dynamic objects. Second, differentiated dynamic feature point rejection strategies are developed for different dynamic objects using the prior information, depth information, and epipolar geometry method. Thus, the localisation and mapping accuracy of the SEG-SLAM system is improved. Finally, the rejection results are fused with the depth information, and a static dense 3D mapping without dynamic objects is constructed using the Point Cloud Library. The SEG-SLAM system is evaluated using public TUM datasets and real-world scenarios. The proposed method is more accurate and robust than current dynamic visual SLAM algorithms.

Technical reports indicate that wireless and mobile devices will account for 71% of all IP traffic by 2022, an increase of 19% over four years. This increase is related to advances in wireless data communication technologies. Wireless networks have become one of the most important ways to connect devices to the Internet, therein improving productivity and encouraging information sharing. IEEE 802.11, known as Wi-Fi, has become the main standard for wireless local area networks. The most important metrics for measuring the quality of Wi-Fi are delay, jitter, and packet loss. Packet loss occurs when one or more packets fail to reach their destination and can occur for a variety of reasons. Packet loss influences the user's perceived quality of applications over Wi-Fi networks, mainly multimedia and real-time applications. The availability of accurate models for packet loss in Wi-Fi networks enables the development of more efficient methods for performance analysis and network design, as well as better computational simulations. Modeling packet loss in such networks presents a major challenge because packets may be lost for many different reasons, including signal attenuation, noise, multipathing, signal refraction, thermal noise, competition for media access and buffer issues. In this paper, we provide an overview of the causes of packet loss and a comprehensive survey of the available models for packet loss in Wi-Fi networks. The potential benefits of the survey are: (i) the systematic presentation of available packet loss models for Wi-Fi networks, their parameters, and respective packet loss rate evaluation, (ii) comparison of models considering validation scenarios and input parameters, and (iii) description of open issues and future research directions. We hope that our analysis will help researchers understand the most important characteristics of the packet loss process in Wi-Fi networks and the strengths and weaknesses of the main packet loss models.

Visual Simultaneous Localization and Mapping (VSLAM) is a key technique that enables autonomous systems to localize themselves and incrementally build a map of an unknown environment using only visual information. Despite its importance, conventional VSLAM systems frequently experience drift and tracking failures in complex environments, which limits their overall effectiveness. Multi-camera systems have enhanced VSLAM accuracy by incorporating diverse viewpoints. However, they generally rely on synchronous data capture, restricting their applicability in multi-sensor setups where asynchronous data acquisition is necessary. To address this limitation, a continuous-time Asynchronous Multi-Camera SLAM (AMC-SLAM) framework is proposed, utilizing sparse Gaussian process regression. The method integrates outlier removal, continuous-time trajectory optimization, and multi-view loop closing to achieve robust pose estimation. Combining Gaussian process interpolation with bundle adjustment strengthens inter-camera data correlation and reduces the number of state variables, while the derived analytical Jacobians enhance optimization efficiency. Additionally, online estimation of multi-camera extrinsic parameters is incorporated to improve the system’s generality. Experimental results on the AMV-Bench dataset [1] demonstrate an absolute translation error of less than 0.5% over a 10 km trajectory, indicating significant improvements in accuracy over existing stereo and multi-camera SLAM systems. The method also exhibits robustness and generalizability on the New College Dataset [2] and in real-world scenarios. These findings underscore the potential of AMC-SLAM for high-precision, robust applications in challenging environments.

The indoor Visual Simultaneous Localization And Mapping (V-SLAM) dataset with various acquisition modalities has been created to evaluate the impact of acquisition modalities on the Visual SLAM algorithm’s accuracy. The dataset contains different sequences acquired with different modalities, including RGB, IR, and depth images in passive stereo and active stereo modes. Each sequence is associated with a reference trajectory constructed with an Structure From Motion (SFM) and Multi View Stereo (MVS) library for comparison. Data were collected using an intrinsically calibrated Intel RealSense D435i camera. The RGB/IR and depth data are spatially aligned, and the stereo images are rectified. The dataset includes various areas, some with low brightness, with changes in brightness, wide, narrow and texture.

Visual simultaneous localization and mapping (SLAM) in dynamic environments has received significant attention in recent years, and accurate segmentation of real dynamic objects is the key to enhancing the accuracy of pose estimation in such environments. In this study, we propose a visual SLAM approach based on ORB-SLAM3, namely dynamic multiple object tracking SLAM (DMOT-SLAM), which can accurately estimate the camera’s pose in dynamic environments while tracking the trajectories of moving objects. We introduce a spatial point correlation constraint and combine it with instance segmentation and epipolar constraint to identify dynamic objects. We integrate the proposed motion check method into DeepSort, an object tracking algorithm, to facilitate inter-frame tracking of dynamic objects. This integration not only enhances the stability of dynamic features detection but also enables the estimation of global motion trajectories for dynamic objects and the construction of object-level semi-dense semantic maps. We evaluate our approach on the public TUM, Bonn, and KITTI dataset, and the results show that our approach has a significant improvement over ORB-SLAM3 in dynamic scenes and performs better compared to other state-of-the-art SLAM approaches. Moreover, experiments in real-world scenarios further substantiate the effectiveness of our approach.

Distilled representation using patch-based local-to-global similarity strategy for visual place recognition

The advent of 5G technology promises a paradigm shift in the realm of telecommunications, offering unprecedented speeds and connectivity. However, the efficient management of traffic in 5G networks remains a critical challenge. It is due to the dynamic and heterogeneous nature of network traffic, including bursty patterns, varying user behaviors, and diverse applications, all of which demand highly accurate and adaptable prediction models to optimize network resource allocation and management. This dissertation investigates the intricate domain of traffic prediction within 5G networks, addressing the specific challenges posed by both massive machine type communication (mMTC) networks and 5G cellular networks. The first segment of this research focuses on mMTC networks, where the event-driven and bursty nature of traffic patterns poses a formidable obstacle to accurate prediction. Forecasting bursty traffic in such environments is a non-trivial task due to the inherent randomness of events, and these challenges intensify within live network environments. Consequently, there is a compelling imperative to design an efficient and agile framework capable of assimilating continuously collected data from the network and accurately forecasting bursty traffic in mMTC networks. The first section of this dissertation is an attempt to addresses these challenges by presenting a machine learning-based framework tailored for forecasting bursty traffic in multi-channel slotted ALOHA networks. We develop a new low-complexity online prediction algorithm that dynamically updates the states of the long-term short-term memory (LSTM) network by leveraging frequently collected data from the mMTC network. Moreover, to evaluate the performance of the proposed framework, we synthesized a realistic mMTC traffic considering both uniform and bursty traffic patterns. In this setup, we considered a single base station and thousands of devices organized into groups with distinct traffic-generating characteristics. Transitioning to the realm of 5G cellular networks, we explore the efficacy of convolutional neural network (CNN)-LSTM and convolutional LSTM (ConvLSTM) models for traffic prediction. Building upon insights from the preceding section, we integrate the proposed live prediction algorithm into these models. Results demonstrate a notable enhancements in prediction accuracy and computational efficiency, signifying a promising avenue for traffic management in 5G cellular networks. Moreover, we study the performance of the proposed live prediction algorithm under the various data collection scenarios. In the final section, we introduce an innovative compression framework to mitigate the data transfer overhead between base stations and centralized nodes. We propose a novel user-specific asymmetric autoencoder (AE)-based data compression framework tailored for data transfer in 5G cellular networks. Leveraging user-specific local AE models and a centralized joint decoder, our framework aims to efficiently compress traffic data while preserving the reconstruction accuracy. In the proposed framework, we utilize a simplified FFNN models in local AEs and CNN layers in the centralized decoder to simultaneously decode the data of all cells by leveraging the spatio-temporal correlations in traffic patterns. Simulation results and complexity analysis highlight the superiority of our proposed live prediction algorithm in terms of both accuracy and computational efficiency, making it well-suited for time-critical live scenarios. In the first section, simulations conducted on synthesized mMTC traffic demonstrate the remarkable accuracy of our machine learning approach in long-term predictions compared to traditional methods, all while imposing minimal additional processing load on the system. The application of our proposed live prediction algorithm to forecast cellular traffic in the second section reveals its superior robustness under both synchronous and asynchronous data gathering scenarios, outperforming traditional methods. Moreover, in asynchronous data gathering scenarios, our algorithm demonstrates the potential to halve the required bandwidth for reporting traffic statistics, illustrating another advantageous aspect of the proposed algorithm. Finally, the performance evaluation of our proposed data compression method on the Telecom Italia dataset in the third section underscores the effectiveness of our approach, achieving superior performance compared to symmetric universal AEs. Furthermore, our framework exhibits reduced complexity, positioning it as a promising solution for practical applications in 5G networks. In summary, this dissertation presents novel methodologies and frameworks aimed at tackling the multifaceted challenges of traffic prediction within diverse 5G network environments. Through the integration of advanced prediction algorithms with innovative data compression techniques, the proposed solutions pave the way for resilient and efficient traffic management in 5G networks, offering promising avenues for future research and implementation.

Mobile Edge Computing (MEC) in 5G networks has emerged as a promising technology to enable efficient and low-latency services for mobile users. In this paper, we present a novel synthetic data generation approach tailored for evaluating MEC in 5G networks. Our methodology incorporates resource-efficient techniques to generate realistic synthetic datasets that capture the spatio-temporal patterns of mobile traffic and user behavior. By leveraging advanced modeling techniques, including multi-head attention and bidirectional LSTM, we accurately model the complex dependencies in the data while optimizing computational resources. The proposed synthetic data generator enables the creation of diverse datasets that closely resemble real-world scenarios, facilitating the evaluation of MEC performance and optimizing resource utilization. Through extensive experiments and evaluations, we demonstrate the effectiveness of our approach in enabling accurate assessments of MEC in 5G networks. Our work contributes to the field by providing a robust methodology for synthetic data generation specifically tailored for MEC evaluation, addressing the need for resource-efficient evaluation frameworks in the context of emerging technologies. The results of our study provide valuable insights for the design and optimization of MEC systems in real-world deployments.

Lift3Dreamer: Boosting Text-Driven Novel View Synthesis via Lifted 3D Inpainting Model from Single Images.

Vehicle viewpoint estimation from monocular images is a crucial component for autonomous driving vehicles and for fleet management applications. In this paper, we make several contributions to advance the state-of-the-art on this problem. We show the effectiveness of applying a smoothing filter to the output neurons of a Convolutional Neural Network (CNN) when estimating vehicle viewpoint. We point out the overlooked fact that, under the same viewpoint, the appearance of a vehicle is strongly influenced by its position in the image plane, which renders viewpoint estimation from appearance an ill-posed problem. We show how, by inserting in the model a CoordConv layer to provide the coordinates of the vehicle, we are able to solve such ambiguity and greatly increase performance. Finally, we introduce a new data augmentation technique that improves viewpoint estimation on vehicles that are closer to the camera or partially occluded. All these improvements let a lightweight CNN reach optimal results while keeping inference time low. An extensive evaluation on a viewpoint estimation benchmark (Pascal3D+) and on actual vehicle camera data (nuScenes) shows that our method significantly outperforms the state-of-the-art in vehicle viewpoint estimation, both in terms of accuracy and memory footprint.

This paper explores the feasibility of leveraging concepts from deep reinforcement learning (DRL) to enable dynamic resource management in Wi-Fi networks implementing distributed multi-user MIMO (D-MIMO). D-MIMO is a technique by which a set of wireless access points are synchronized and grouped together to jointly serve multiple users simultaneously. This paper addresses two dynamic resource management problems pertaining to D-MIMO Wi-Fi networks: (i) channel assignment of D-MIMO groups, and (ii) deciding how to cluster access points to form D-MIMO groups, in order to maximize user throughput performance. These problems are known to be NP-Hard and only heuristic solutions exist in literature. We construct a DRL framework through which a learning agent interacts with a D-MIMO Wi-Fi network, learns about the network environment, and is successful in converging to policies which address the aforementioned problems. Through extensive simulations and on-line training based on D-MIMO Wi-Fi networks, this paper demonstrates the efficacy of DRL in achieving an improvement of 20% in user throughput performance compared to heuristic solutions, particularly when network conditions are dynamic. This work also showcases the effectiveness of DRL in meeting multiple network objectives simultaneously, for instance, maximizing throughput of users as well as fairness of throughput among them.

The performance of Mobile Broadband (MBB) services of Fourth Generation (4G) and Third Generation (3G) mobile networks over urban morphology is studied in Malaysia based on experimental measurements of drive test data. The aim of this study is to provide a roadmap for service providers to establish a realistic plan for future Fifth Generation (5G) networks. This work is a continuation of our previous work for the scope of rural areas in Malaysia. The MBB measurement data have been gathered through drive tests conducted in the urban areas of four states throughout Malaysia (namely, Klang Valley/Selangor, Johor, Sarawak and Sabah) to characterise and analyse MBB performances. The gathered data are from the cities, highways and federal roads of the chosen states, and encompasses three main Mobile Network Operators (MNOs). Data has been collected in a time span of 2 months, from January to February, using the Samsung Galaxy S6 smartphone handsets. Four MBB Key Performance Indicators (KPIs) are considered in this study (coverage, latency, satisfaction and speed) for two MBB services (web browsing and video streaming). The measurement data for characterising the performance of each MBB service has been collected using a dedicated smartphone handset. YouTube videos with 720p and 1080p resolutions have been sequentially streamed to assess the performance of MBB video-streaming services. Three distinct websites (Google, Instagram and mStar) have been accessed to evaluate the performance of MBB web-browsing services. The experimental methodology of this study integrates several diversified elements including four different urban states, four distinct KPIs, three main MNOs, two MBB services and two radio networks (4G and 3G), which are both accessible by the smartphones when available to mimic real-world scenarios. The results of this study reveal that the performance of 4G radio networks is generally superior to that of 3G. For instance, 4G networks achieved a vMOS score of more than 3 for both MBB video-streaming and web-browsing services, while 3G networks scored less than 3 across all four study areas. The analysed experimental results confirmed that compared to 3G networks, 4G technology presents an enhancement factor of up to 1.6 and 4.2 in download speed when streaming a video and browsing a web page, respectively. The study outcomes can contribute to the efficient planning of non-standalone (NSA) 5G networks in Malaysia where 5G networks will be aided by existing 4G infrastructures. Analysing the 4G coverage performance is the first step towards deciding the deployment rate of NSA 5G in Malaysia.

The performance of visual SLAM in complex, real-world scenarios is often compromised by unreliable feature extraction and matching when using handcrafted features. Although deep learning-based local features excel at capturing high-level information and perform well on matching benchmarks, they struggle with generalization in continuous motion scenes, adversely affecting loop detection accuracy. Our system employs a Model-Agnostic Meta-Learning (MAML) strategy to optimize the training of keypoint extraction networks, enhancing their adaptability to diverse environments. Additionally, we introduce a coarse-to-fine feature tracking mechanism for learned keypoints. It begins with a direct method to approximate the relative pose between consecutive frames, followed by a feature matching method for refined pose estimation. To mitigate cumulative positioning errors, DK-SLAM incorporates a novel online learning module that utilizes binary features for loop closure detection. This module dynamically identifies loop nodes within a sequence, ensuring accurate and efficient localization. Experimental evaluations on publicly available datasets demonstrate that DK-SLAM outperforms leading traditional and learning-based SLAM systems, such as ORB-SLAM3 and LIFT-SLAM. DK-SLAM achieves 17.7% better translation accuracy and 24.2% better rotation accuracy than ORB-SLAM3 on KITTI and 34.2% better translation accuracy on EuRoC. These results underscore the efficacy and robustness of our DK-SLAM in varied and challenging real-world environments.

For road maintenance up-to-date information about road conditions is important. Such information is currently expensive to obtain. Specially equipped measuring vehicles have to perform surface scans of the road, and it is unclear how to automatically Ąnd faulty sections in these scans. This research solves the problem by stereo vision with cameras mounted behind the windshield of a moving vehicle so that the system can easily be integrated into a large number of vehicles. The stereo images are processed into a depth map of the road surface. In a second step, color images from the cameras are combined with the depth map and are classified by a convolutional neural network. It is shown that the developed system is able to Ąnd defects that require knowledge about surface deformations. These defects could not have been found on monocular images. The images are taken at usual driving speed.

With the development of computer science, there is an increasing demand on the object recognition in stereo images. As a binocular image pair contains larger and more complicated information than a monocular image, the stereo vision analysis has been a difficult task. Therefore how to extract the region of user’s interest is a vital step to reduce the data redundancy and improve the robustness and reliability of the analysis. The original stereo sequences used in the paper are obtained from two parallel video cameras mounted on a vehicle driving in a residential area. This paper targets the problem of data segmentation of those stereo images. It proposes a set of algorithms to separate the foreground from the complicated changing background. Experiments show that the whole process is fast and efficient in reducing the data redundancy, and improves the overall performance for the further obstacle extraction.KeywordsStereo ImageData SegmentationData RedundancyLeast Square Fitting