Abstract In the autonomous driving environment, object instances in an image can be affected by various factors such as camera, driving state, weather, and system component. However, the deep learning‐based vision systems are vulnerable to perturbation, which contains noise. Thus, robust object detection under harsh autonomous‐driving environments is a more difficult than the generic situation. In this paper, it is found that not only the accuracy, but also the speed of the non‐maximum suppression‐based detector can be degraded under harsh environments. Therefore, object detection is handled under a harsh situation with adversarial mechanisms such as adversarial training and adversarial defence. Adversarial defence modules are designed to improve robustness in feature extraction level and define perturbations under a harsh environment for training object detectors to improve the robustness of the model's decision boundary. The proposed adversarial defence and training mechanisms improve the object detector in both accuracy and speed. The proposed method shows a 43.7% mean average precision for the COCO2015 dataset in generic object detection and 39.0% mean average precision for the BDD100K dataset in a driving environment. Furthermore, it achieves a real‐time capability of 23 frames per second.

To better understand early brain development in health and disorder, it is critical to accurately segment infant brain magnetic resonance (MR) images into white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF). Deep learning-based methods have achieved state-of-the-art performance; h owever, one of the major limitations is that the learning-based methods may suffer from the multi-site issue, that is, the models trained on a dataset from one site may not be applicable to the datasets acquired from other sites with different imaging protocols/scanners. To promote methodological development in the community, the iSeg-2019 challenge (http://iseg2019.web.unc.edu) provides a set of 6-month infant subjects from multiple sites with different protocols/scanners for the participating methods. T raining/validation subjects are from UNC (MAP) and testing subjects are from UNC/UMN (BCP), Stanford University, and Emory University. By the time of writing, there are 30 automatic segmentation methods participated in the iSeg-2019. In this article, 8 top-ranked methods were reviewed by detailing their pipelines/implementations, presenting experimental results, and evaluating performance across different sites in terms of whole brain, regions of interest, and gyral landmark curves. We further pointed out their limitations and possible directions for addressing the multi-site issue. We find that multi-site consistency is still an open issue. We hope that the multi-site dataset in the iSeg-2019 and this review article will attract more researchers to address the challenging and critical multi-site issue in practice.

This study proposes an effective delay‐aware packet forwarding (DAPF) for driving safety and efficiency in vehicular networks. Vehicular ad hoc networks have been an emerging technology for vehicular communication for the last few decades, but still, it has many challenging issues such as on‐time dissemination of message at an emergency situation (e.g. accident and obstacle) to the vehicles having the same route to their destinations. This on‐time dissemination can prevent further collision of vehicles and road traffic congestion. In this study, the authors propose an effective way of selecting the processing position of a message among a cluster head, road‐side unit (RSU), and vehicular cloud, on the basis of total delivery time and cost. They further show that this effective selection and on‐time dissemination helps the upcoming vehicles to select an appropriate route to their destinations. Through simulation results, it is shown that their DAPF outperforms other schemes in terms of packet delivery time.

Accurate segmentation of infant brain magnetic resonance (MR) images into white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) is an indispensable foundation for early studying of brain growth patterns and morphological changes in neurodevelopmental disorders. Nevertheless, in the isointense phase (approximately 6-9 months of age), due to inherent myelination and maturation process, WM and GM exhibit similar levels of intensity in both T1-weighted (T1w) and T2-weighted (T2w) MR images, making tissue segmentation very challenging. Despite many efforts were devoted to brain segmentation, only few studies have focused on the segmentation of 6-month infant brain images. With the idea of boosting methodological development in the community, iSeg-2017 challenge (http://iseg2017.web.unc.edu) provides a set of 6-month infant subjects with manual labels for training and testing the participating methods. Among the 21 automatic segmentation methods participating in iSeg-2017, we review the 8 top-ranked teams, in terms of Dice ratio, modified Hausdorff distance and average surface distance, and introduce their pipelines, implementations, as well as source codes. We further discuss limitations and possible future directions. We hope the dataset in iSeg-2017 and this review article could provide insights into methodological development for the community.

The localised active contour framework has been widely used for image segmentation because it provides reliable results for inhomogeneous images. However, its computational complexity remains an issue. In this study, the authors introduce a fast algorithm based on the localised active contour framework. A key concept of the proposed algorithm is its consideration of the curve evolution based on the speed function only at active points that change across time, rather than at all points in a narrow band. This approach reduces computational time in the localised active contour. The authors additionally propose a modified speed function to address inhomogeneous image segmentation. The experimental results demonstrate significant advantages of the proposed method over existing methods, both in terms of computational efficiency and segmentation accuracy, for homogeneous and inhomogeneous images.

This study investigates the cooperative multicast of scalable video over dedicated relay‐based cellular networks. Different layers in a scalable video tend to have different transmission quality requirements, and legacy cellular network has limitations in providing homogeneous service quality within its coverage area. The authors propose two realistic cooperative transmission schemes with a capacity performance analysis for scalable video over dedicated relay‐based cellular networks. The proposed transmission schemes with multiple antennas at an evolved node B and the dedicated relay utilise the combinations of beamforming and space‐time coding in order to enhance the multicast performance of scalable video. The proposed schemes ensure that the received base layer (BL) of scalable video guarantees the minimum video quality to all user terminals throughout the cell coverage with given outage probability, while the terminals inside the relay coverage can achieve better video quality by receiving both the BL and additional enhancement layer data according to the received channel condition. The antennas configuration and the video type determine which multicast scheme to choose.

The recently emerging location-based services in long-term evolution (LTE) systems require the accurate and efficient tracking of mobile users. An optimal information-theoretic framework is developed for tracking area update (TAU) for next-generation LTE cellular systems. Shannon's entropy is used to characterise the location uncertainty of the mobile users. Based on this entropy-based tracking framework, two practical location management schemes, Bayesian-based TAU and entropy-coding based TAU, are proposed. The proposed schemes capture the users' mobility patterns online and perform profile-based paging to optimise the TAU cost. Of the two proposed schemes, the Bayesian-based TAU operates as an independently identically distributed process and improves the paging cost with less storage and computational overhead, whereas the entropy-coding-based TAU using the Lempel–Ziv strategy asymptotically minimises both the update and paging costs with higher storage and computational overheads than the Bayesian-based one. There is some trade-off between the update/paging costs and storage/computational overheads in the two proposed schemes. The simulation results demonstrate that both proposed schemes outperform the existing comparable schemes for LTE systems in all of the performance metrics.

This paper designs a distributed Transmission Opportunity (TXOP) adaptation algorithm for IEEE802.11e Enhanced Distributed Channel Access (EDCA). Each node measures its throughput in a window and compares it with a target value. If the measured throughput is higher than the target value, the node reduces its TXOP, otherwise if the measured value is less than the target throughput, the node increases its TXOP. We show that the target throughput can be achieved in a globally stable manner.

In this paper, we propose a QoS tracking algorithm to support different QoS requirements to accommodate delay-sensitive and throughput-sensitive traffics in wireless mesh networks. The proposed scheme is carried out in a totally distributed manner by adjusting the minimum contention window size. Moreover, we investigate whether we can satisfy all QoS requirements in a stable manner through stability analysis. The simulation results show that the proposed algorithm can satisfy both average delay bound and minimum throughput requirements without sacrifice of total system efficiency.

Femtocells are being deployed for coverage extension and higher data rates. However, interference problem between macrocell and femtocells should be solved before massive deployment. In this paper, we analyze the possible cases of frequency resource allocation in the OFDMA femtocell systems using Fractional Frequency Reuse (FFR) in order to minimize the interference between both cells. Under the assumed pre-allocation of macrocell users with the FFR, the allocable cases of sub-bands for the femtocells are investigated. Efficient and practical guidelines for femtocell's resource allocation are suggested with the consideration of macrocell users having a priority over femtocell users and of enhancing overall throughputs. Simulation results show the enhanced throughputs of both macrocell users and combined total users through practical interference-avoidance guidelines.

Femtocell is one of promising key technologies to increase the efficiency and coverage of next generation broadband wireless networks. However, there exists interference problem between femtocells and macrocell due to co-channel operation. In this paper, we propose an efficient frequency resource allocation scheme for newly deployed femtocells in the legacy OFDMA networks in order to maximize the throughput of both macrocell/femtocell users without severe degradation to the existing macrocell users. Under the pre-allocated frequency bands for the macrocell users with the sectored-FFR is given, the proposed scheme allocates sub-bands to the femtocells efficiently to consider macrocell users having a priority over femtocell users. Simulation results show that the proposed one obtains higher total/edge throughputs than compared schemes.

An information-theoretic, optimal framework is developed for tracking the residents in a context-aware heterogeneous smart environment. The framework envisions that each individual sensor system operates fairly independently and does not require public knowledge of individual topologies. The resident-tracking problem is formulated in terms of a new concept of weighted entropy. The framework is truly universal and provides an optimal, online learning and prediction of inhabitant's movement (location) profiles from the symbolic domain. The overall optimal tracking in heterogeneous smart homes is proved to be an NP-complete problem, and a greedy heuristic for near-optimal tracking is proposed. The concept of asymptotic equipartition property is also explored to predict the inhabitant's most likely path segments (comprising coverage areas of different sensor systems) with very good accuracy. Successful prediction helps in on-demand operations of automated indoor devices along the inhabitant's future paths and locations, thus providing the necessary comfort at a near-optimal cost. Simulation results on a typical smart home corroborate a high prediction success of ∼91%, thereby providing sufficient resident-comfort (≥7 in the scale of 10) while reducing the daily energy consumption and manual operations to less than one-third of its original values.

Abstract Rapid penetration of smart wireless devices and enormous growth of wireless communication technologies has already set the stage for deployment of wireless sensor networks (WSNs). While these small sensor nodes are often considered as the future of wireless communications, they also suffer from energy constraints. On the other hand, with increasing demand for real‐time services in next generation wireless networks, quality‐of‐service (QoS)‐based routing has emerged as an interesting research topic. Naturally offering some QoS‐guarantee in sensor networks raises significant challenges. The network needs to cope up with battery‐constraints, while providing precise QoS (end‐to‐end delay and bandwidth requirement) guarantee. More precisely, designing such QoS‐protocols, optimizing multiple objectives, is computationally intractable. Based on the multi‐objective genetic algorithm (MOGA), in this paper we propose a QoS‐based energy‐efficient sensor routing (QuESt) protocol that determines application‐specific, near‐optimal sensory‐routes demand, by optimizing multiple QoS parameters, (end‐to‐end delay and bandwidth requirements) and energy consumption. Simulation results demonstrate that the proposed protocol is capable of discovering a set of QoS‐based, near‐optimal routes, even with imprecise network information. Copyright © 2007 John Wiley & Sons, Ltd.

Emerging applications like video surveillance, telemedicine and traffic monitoring requires multimedia transmission over wireless sensor networks. Offering some precise QoS-guarantee (e.g.end-to-end delay and bandwidth requirement) for wireless multimedia over already battery-constrained sensor networks raises significant challenges. This demands for a suitable sensory MAC protocol capable of achieving application- specific QoS. In this paper we have proposed a new QoS-based sensory MAC protocol, which not only adapts to application-oriented QoS, but also attempts to conserve energy without violating QoS-constraints. Performance modelling, analysis and simulation results demonstrate that the proposed protocol is capable of providing lower delay and better throughput, at the cost of reasonable energy consumption, in comparison to other existing sensory MAC protocols.

We explore a methodology to divide an application's data into several categories and send them through different IP differentiated services vices (DiffServ) grades. The issues are motivated by a futuristic infrastructure wherein the Internet service provider charges different prices to packets going through different service grades. This paper discusses how a subscriber of the DiffServ can assign different grades to real-time video packets on the basis of the packets' relative importance under the constraint of communication expenses. For relative DiffServ, we present an adaptive packet forwarding mechanism to preserve the performance ratio among service grades persistently over different time scales. Through experiments we observe how this packet forwarding mechanism improves the the quality of the application that maps the packets into the different service grades. In the experiment the end-to-end performance is evaluated by the error resilient packet video transmission using ITU-T H.263+ codec over a simulated Diffserv network.

This paper presents a futuristic framework for quality-of-service (QoS) mapping between practically categorized packet video and relative differentiated service (DiffServ or DS) network employing unified priority index and adaptive packet forwarding mechanism under a given pricing model (e.g., DiffServ level differentiated price/packet). Video categorization is based on the relative priority index (RPI), which represents the relative preference per each packet in terms of loss and delay. We propose an adaptive packet forwarding mechanism for a DiffServ network to provide persistent service differentiation. Effective QoS mapping is then performed by mapping video packets onto different DiffServ levels based on RPI. To verify the efficiency of proposed strategy, the end-to-end performance is evaluated through an error resilient packet video transmission using ITU-T H.263+ codec over a simulated DiffServ network. Results show that the proposed QoS mapping mechanism can exploit the relative DiffServ advantage and result in the persistent service differentiation among DiffServ levels and the enhanced end-to-end video quality with the same pricing constraint.

In this research, we investigate quality of service (QoS) interaction provisioning between video applications and components of the DiffServ network. QoS interaction is performed through the mapping of video packets based on the relative quality index (RQI), which represents the relative preference in terms of loss and delay, onto the adaptive packet forwarding mechanism in a DiffServ network. To verify the efficiency of the proposed strategy, the end-to-end performance is evaluated through error resilient packet video transmission over a simulated DiffServ network. Results show that the proposed QoS interaction framework results in an enhanced end-to-end video quality at the same cost constraint.