Monocular depth estimation is gaining attention in computer vision for autonomous driving due to its cost-effectiveness and versatility. Recent works have used self-supervised learning to tackle this problem. However, they overlook crucial challenges for real-world deployment, such as data privacy, network consumption, computational cost, and unstable connectivity. Recent studies have shown the potential benefits of federated learning in addressing these challenges. Thus, we introduce FedSCDepth, a novel method that combines federated learning and deep self-supervision to efficiently learn monocular depth estimators with performance comparable to the state-of-the-art. Our evaluation using the KITTI dataset’s Eigen’s Split reveals that FedSCDepth achieves high efficacy, with a test loss below 0.13. It also demonstrates impressive efficiency, requiring only an average of 1.5k training steps and up to 0.415 GB of weight data transfer per autonomous vehicle in each round.

Recent investigations in computer vision for autonomous vehicles have focused on depth estimation from images, owing to its cost-efficiency and adaptability. Monocular depth estimation, using a single camera, is notable for its adaptability compared to binocular techniques that require two fixed cameras. Sophisticated methodologies employ self-supervised deep neural networks, while latest research proposes the use of federated learning to tackle crucial challenges for autonomous driving, such as data privacy, network usage, computation distribution, and connectivity robustness. Nevertheless, continual learning presents an additional challenge for both centralized and federated training, as updating models with novel datasets may induce the forgetting of previously acquired samples, thereby diminishing the model’s versatility. Meanwhile, recent research indicates that experience replay strategies can be used to alleviate forgetting in autonomous driving use cases. We introduce ERFedSCDepth, an innovative approach amalgamating experience replay, federated learning, and deep self-supervision to allow the training of monocular depth estimators with high effectiveness and efficiency. Assessment using KITTI and DDAD datasets shows the efficacy of our approach, achieving enhanced continual learning performance over state-of-the-art baseline at standard depth metrics.

Incentive mechanisms are increasingly needed in opportunistic networks that contain nodes with selfish behavior. For this, credit-based mechanisms need a virtual bank (central entity) to promote the incentive. However, the existence of this central entity in an opportunistic network may be challenging. Therefore, we propose a credit incentive mechanism in this paper. The main contribution is that the mechanism does not use a virtual bank to distribute credits (reward for forwarding messages) but rather a decentralized approach. In addition, we propose mathematical modeling to represent the collection and distribution of credits to avoid Edge Insertion attacks in some instances. Finally, the proposed mechanism was evaluated through simulation using real mobility traces and different routing protocols and compared its performance with the RELICS incentive mechanism. The obtained results show that the proposed mechanism is promising in diminishing the occurrence of selfish nodes.

Increasing the availability of applications is a way to improve service delivery to final consumers. This paper proposes an architecture based on fog computing and attends intelligent transportation system applications with stricter latency requirements. We also present an algorithm based on autoregressive neural networks and adaptive elasticity thresholds to guarantee applications' availability. The proposed algorithm aims to decrease the number of lost requests coming from the terminal layer. Based on workload, this algorithm analyzes the server resources in a fog-based solution and indicates the best time to increase or decrease the infrastructure. This paper also presents an experimental assessment of different scenarios to manage applications' elasticity using several techniques and policies. Finally, we compare the proposed algorithm with state-of-the-art, and its results showed a reduction of 17.8% in the number of requests lost than the others.

In urban mobility scope, one of the main current challenges is fast and localized information gathering for decision making. In this sense, social media can be an information extraction source for several tasks, such as traffic control. However, these data have to be well classified in order to ensure that only relevant information is used. Particularly in Portuguese-speaking countries, there are not many works with this approach. Thus, we designed a methodology to identify traffic information from social media using modern techniques for classification. In addition, we generated a dataset, by extracting and manually labeling tweets in Portuguese language, and used deep learning techniques to classify these messages with respect to the traffic domain. Finally, we perform experiments employing different combinations of deep learning architectures for representation and classification, obtaining results of more than 95% on average for accuracy and precision metrics.

Motivated by Smart City applications and services, this article presents a novel approach to identifyinguser communitiesin communication networks. We define user communities as groups of users that share common mobility features over spatio-temporal scales of arbitrary length, such as time spent in certain locales, mobility speed, and time between consecutive movements. We describe our user community identification framework in detail including how mobility features can be extracted from real mobility traces (as examples of unlabelled data) and synthetic mobility records (as examples of labeled data). We present results obtained when using our approach in four distinct mobility scenarios represented by both unlabeled and labeled datasets. We also introduce a new validation methodology that uses image-based similarity metrics in order to assess the quality of identified communities. Our results show that the proposed approach significantly increases similarity between users within the same community as well as dissimilarity between users in different communities. We also demonstrate that the proposed user community identification approach yields significant increase in contact time amongst users belonging to the same community when compared to the average contact time when not considering community structures.

Using Edge Computing approaches is a way to improve the Intelligent Transportation Systems (ITS) applications in smart cities. This article proposes a case study of an ITS system with a Mobile Edge Computing (MEC) architecture using the platform for time-sensitive ITS applications. The platform aims to support ITS applications with strict requirements for latency and requests loss. We evaluated an ITS application in three different scenarios: in the cloud using a 4G connection, in the cloud using broadband, and in MEC. We compare scenarios using quantitative metrics such as requests per second, average response time, and request loss rate. The experimental results indicate that MEC has a lower loss of requests and it can work with a greater number of requests per second, which makes it a good solution for ITS applications.

With the growth of large cities, it becomes necessary to overcome several challenges related to infrastructure. In this context, urban mobility, including transportation systems, is a key topic. It is known that more efficient traffic management makes cities more efficient as a whole, improving the quality of life of their residents and tourists. A common problem that arises concerns traffic jams, encompassing their detection and management. With proper traffic jam detection, it is possible to implement traffic management methods to prevent long-term occurrences or even notify drivers to avoid specific locations. Some traffic jam detection solutions use sensors spread across cities, while others are based on sensors in drivers' smartphones. These approaches have limitations related mainly to the information reliability or to the cost and complexity of implementation. In this work, we propose an approach using Machine Learning techniques applied to public data of transportation systems to detect urban traffic jams. The proposed approach is evaluated using the publicly available GPS data from the buses of the Rio de Janeiro/Brazil transportation system and traffic jam events posted on Twitter by the Rio de Janeiro City Hall.

The analysis of vehicular traffic and the precise forecast of the user demands enable the development of intelligent vehicular networks. It is usually necessary a considerable effort to obtain, process, and store the data from sensors in vehicles or city infrastructure before using it in benefit of meeting vehicular application's communication requirements. This work deals with the problem of providing information about the vehicular context for the network infrastructure supporting modern vehicular applications. We want to determine whether machine learning (ML) models could be trained on the network measurements of vehicular applications and used to predict events in the vehicular context, such as traffic jams, accidents, among others. We conduct a proof of concept with a typical scenario of an urban traffic jam consisting of vehicles associated with different vehicular applications using the same network infrastructure. The data generated from the network measurements is used to train and test different supervised ML algorithms. The results show that it is possible to use ML applied to the communication network to obtain the vehicular context. The main gain of our proposal is the evaluation of ML applied to the network communication data to predict the vehicular context. This novel approach has the advantage of using data that can be easily obtained in the network core, providing resources that can be used to meet the requirements of modern vehicular applications of Intelligent Transport Systems efficiently.

Vehicular networks are critical pieces in support of advanced intelligent transportation systems (ITS). These networks are formed by vehicles that can be connected to one another as well as to the infrastructure, and are subject to constant topology changes, disconnections, and data congestion. Each ITS application could have a different set of communication requirements, such as delay, bandwidth, and packet delivery ratio. Meeting these heterogeneous requirements in the complex dynamic environment of vehicular networks is a challenge. This paper develops a new framework for application-driven vehicular networks using 5G network slicing. We present the architecture of the proposed solution and design algorithms for heterogeneous traffic in a dynamic vehicular environment. Our simulations on realistic vehicular scenarios show significant improvements in network performance compared to the state-of-the-art approaches.

In this paper we propose two mechanisms of traffic jams’ detection and control aiming to promote an efficient mobility in urban centers using exclusively the vehicle-to-vehicle approach. These approaches dispenses the existence of vehicular network infrastructure and, therefore, removing the demand of centralized resources. Finally, a performance evaluation using simulated scenarios with traffic jams indicated that the proposed mechanisms were able to reduce the traffic jams, decreasing the average travel time by 30%1This work was partially supported by CAPES (Ministry of Education, Brazil)

Intelligent Transportation System's main objective is to provide sustainable and optimized means of transportation for citizens of urban centers. Identifying the travel modes used by citizens, in real- time, allows these systems to better adapt transportation infrastructure and services according to user needs and possibilities of interaction. Many works have explored the use of machine learning algorithms and ensemble methods, combined with general and domain specific feature extraction techniques, for this task. More recently, several works evaluated the use of deep learning algorithms while none of them has explored the use of Recurrent Neural Networks (RNNs) in conjunction with domain feature extraction to enable the development of flexible and lightweight travel mode detection solutions based on multiple smartphone sensor readings. In this paper, we propose a deep RNN architecture for building online travel mode detection models using Long-Short Term Memory (LSTM) cells and evaluate its performance using real mobility data collected with a wide variety of sensors. The experiments show that the proposed architecture allows the generation of models with high accuracy and lower memory consumption and computation cost than state-of-the-art supervised machine learning (ML) approaches.

Opportunistic networking has been proposed to address episodic connectivity, common in so-called "challenged" or "extreme" networking environments where arbitrarily frequent and long-lived connectivity disruptions are the norm, instead of the exception. Examples of extreme networking environments and applications include interplanetary communication, disaster relief, and emergency response, (semi-)autonomous driving, to name a few. This paper proposes a novel community-based opportunistic routing approach that identifies user communities based on mobility features extracted from real traces of user mobility. The proposed Deep AutoenCoder Community-based Opportunistic Routing protocol, or DACCOR, employs deep learning to identify user communities based on data extracted from user mobility traces and uses user community information to make forwarding decisions in opportunistic networking scenarios. Through extensive simulations, we evaluate DACCOR's performance and show that it outperforms well-known opportunistic forwarding protocols in terms of delivery probability, latency, and communication overhead. We also show that DACCOR's lower communication overhead yields considerable energy efficiency, increasing mobile devices' battery lifetime.

Identifying the travel modes people use on their daily commute, and the purpose of their trips, contributes to a better understanding of their mobility patterns. Therefore, the detection of the travel modes and the prediction of trips, purposes from smartphone sensors data emerged as two research problems in recent years. Most previous works have tackled these problems in separate studies, leading to an unnecessary redundancy with respect to the preprocessing steps taken. In this paper, we propose the use of a single preprocessing algorithm for both problems, using location traces obtained through smartphone sensors. The proposed technique applies multiple preprocessing steps to extract relevant features for both classification tasks and uses automated machine learning methods to identify the best classifiers and its hyperparameter configurations for each one of them. During evaluation experiments conducted with real smartphones data, a maximum accuracy of 88% for travel mode detection and 81% for trip purpose prediction were reached using our technique.

Device-to-Device (D2D) is a technology that allows the direct communication between the units of a Long Term Evolution - Advanced (LTE-A) network, allowing the forms of coverage to be more cost-effective, more efficient and flexible. Although the use of D2D is quite promising, there are still several challenges. Mobility is undoubtedly one of challenges to be considered, specially when we are interested in evaluating the network performance during its occurrence. Thus, this paper proposes a model to evaluate the performance of the D2D mechanism in LTE-A networks, by updating its network parameters in occurrence of mobility. The proposed modeling was implemented through the Python language. A performance evaluation was carried out and important results were obtained in relation to the percentage of users served by the variation of density and range of base station and D2D devices. In addition, an analysis of influence of devices speed variation was carried out.

Context-aware applications in intelligent transportation systems have a growing need for travel mode detection systems. However, few applications allow real-time travel mode detection through the use of smartphones. In this paper, we propose a real-time travel mode detection application based on GPS traces using a data mining technique through which these traces are preprocessed, grouped in motion segments and classified by supervised machine learning algorithms. An application prototype was implemented on the Android platform, used by smartphones, for movement data collection and user travel mode detection using the WEKA API in Java. Finally, to evaluate the performance of the application in a real environment, field tests were carried out with dozens of volunteers in the metropolitan area of Rio de Janeiro. Therefore, 1338 travel mode inferences were obtained by four machine learning techniques and the results were evaluated and compared by the indicators of the confusion matrix. Thus, through the performance evaluation carried out, it was possible to verify that the proposed application is useful for real-time travel mode detection in urban centers.

Content-Centric Networks (CCN) has emerged as an innovative paradigm that modifies the traditional host-centric and address-based communication model. Its application to delay tolerant network scenarios is known to positively impact some important network metrics and characteristics, such as delay and intermittence. Thus, we justify the feasibility of this combination and present DIRESC, a protocol for selection of message relays for data announcement, discovery and retrieval through a scheme of social cooperation between nodes. We implement our proposal using the ONE simulator and compared its performance against CCN broadcasting, PIFP and STCR protocols. Our results show that DIRESC is able to decrease at least 73.3% of data recovery delay and a gain of at least 9.88% in satisfaction of content requests over PIFP. Thus, we present DIRESC as a promising solution for accelerating the process of obtaining data in this type of network.

Opportunistic networks are formed by selforganizing wireless nodes, in which data delivery is not guaranteed, due to intermittent end-to-end connectivity. In order to infer future contact opportunities, a large amount of information extracted from the context of users, such as its social behavior, associated to connectivity metrics have been used to design new forwarding protocols for opportunistic networks. Yet one open issue in this approach is the frequent usage of repeated nodes considered as the most likely to deliver the messages, thus funneling forwarded information towards certain routes containing such nodes overloading them and, therefore, causing a faster consumption of its battery resources. This paper presents SOCLEER, a social-based forwarding protocol which proposal is to relieve the constant demand for nodes considered most popular for forwarding messages in an opportunistic network. To this end, a performance evaluation was fulfilled and the SOCLEER protocol obtained better results than BUBBLE-Rap on forwarded messages load distribution, as well as a battery consumption reduction of preferred nodes when also compared to Epidemic and PROPHET protocols.

This paper presents the URBAN XOR1protocol, an XOR-based flat routing mechanism developed for vehicular ad hoc networks (VANETs) formed in urban scenarios. The paper firstly describes the URBAN_XOR routing principle, which requires reduced knowledge about the set of nodes present in the VANET in order to provide traffic forwarding. Basically, the URBAN XOR protocol introduces the concept of local visibility, prioritizing the insertion of closer neighbors (in number of hops) in the routing tables, and simplifying the management of the frequent network mobility found in VANETs. Then, the performance of the proposed URBAN_XOR protocol is compared through simulation with other topology-based and position-based protocols, characterizing its performance in terms of path availability ratio, end-to-end delay, path length and path duration. The results reveal that URBAN_XOR contributes for the overall network stability, reducing the end-to-end delay due to its ability of generating shorter paths. At the same time, URBAN_XOR exhibits path availability similar to other topology-based protocols, but exhibiting better path duration times.

We present a new analysis for human mobility through real traces captured in a recreational environment. The goal of this analysis is to investigate the motion components, in both a qualitative and quantitative way, and thus, to get a better knowledge of pedestrian mobility behavior. Moreover, some results based upon human mobility captured in a real scenario with GPS equipment are presented. In these captured data, we verified that the speed and acceleration components follow a Normal distribution, while the direction angle variation component and the pause time measure follow a Lognormal distribution. Finally, we show that velocity and direction angle change components of captured scenario have a temporal dependence unlike the random mobility models.

In this paper is proposed a real mobility characterization study for wireless devices. For this, we use mobility metrics and statistic analysis to evaluate the behavior of device movement components.

Mobility models try to represent the movement behavior of devices in mobile ad hoc networks. These models are used in performance evaluation of applications and communication systems, allowing the analysis of the mobility's impact. In this context, two individual mobility models for mobile ad hoc networks are proposed in this paper. These models were based in [C Chaing et al. (1998)] and intend to represent a wider movement capability. Using the proposed models it is possible to move in the same direction, in adjacent directions, to accelerate and to stop, avoiding sharp turns and sudden stops. This way, it is tried a closer representation to the real movement of the users in urban environments and roads. In this paper, the models are described analytically by Markov chains. Besides that, some comparisons between the proposed models, the waypoint model and the random walk model are presented through the use of simulations.