A work-flow which aims at capturing residents’ abnormal activities through the passenger flow of elevator in multi-storey residence buildings is presented in this paper. Firstly, sensors (hall sensor, photoelectric sensor, gyro, accelerometer, barometer, and thermometer) connected with internet are mounted in elevator to collect image and data. Then computer vision algorithms such as instance segmentation, multi-label recognition, embedding and clustering are applied to generalize passenger flow of elevator, i.e. how many people and what kinds of people get in and out of the elevator on each floor. More specifically so-called GraftNet is proposed for fine-grained multi-label recognition task to recognize human attributes(e.g. gender, age, appearance, and occupation). Thirdly, based on the passenger flow data, anomaly detection of unsupervised learning is hierarchically applied to detect abnormal or even illegal activities of the residents. Meanwhile, based on manual reviewed data, Catboost algorithm is implemented for multi-classification task. Experiment shows the work-flow proposed in this paper can detect the anomaly and classify different categories well.
Authors: Shuai Zhu (Shanghai Elevator Media Information Co., Ltd.), Chunhua Jia (Shanghai Elevator Media Information Co., Ltd.), Wenhai Yi (Shanghai Elevator Media Information Co., Ltd.), Yu Wu (Shanghai Elevator Media Information Co., Ltd), Zhuang Li (Shanghai Elevator Media Information Co., Ltd.), Leilei Wu (Shanghai Elevator Media Information Co., Ltd.),
Hide Authors & Abstract

With the development of the new generation of information technology, traditional factories are gradually transforming into smart factories. How to meet the low-latency requirements of task processing in smart factories so as to improve factory production efficiency is still a problem to be studied. For real-time tasks in smart factories, this paper proposes a resource sched-uling architecture combined with cloud and fog computing, and establishes a real-time task delay optimization model in smart factories based on the ARQ (Automatic Repeat-request) protocol. For the solution of the optimi-zation model, this paper proposes the GSA-P (Genetic Scheduling Arithme-tic With Penalty Function) algorithm to solve the model based on the GSA (Genetic Scheduling Arithmetic) algorithm. Simulation experiments show that when the penalty factor of the GSA-P algorithm is set to 6, the total task processing delay of the GSA-P algorithm is about 80% lower than that of the GSA-R(Genetic Scheduling Arithmetic Reasonable) algorithm, and 66% lower than that of the Joines& Houck method algorithm; In addition, the simulation results show that the combined cloud and fog computing method used in this paper reduces the total task delay by 18% and 7% compared with the traditional cloud computing and pure fog computing methods, respectively.
Authors: Ming-Tuo Zhou (Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences), Tian-Feng Ren (Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences), Zhi-Ming Dai (Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences), Xin-Yu Feng (Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences),
Hide Authors & Abstract

Medical sensor networks provide rich contextual information and alerting mechanisms with continuous monitoring. Broadcast is an important communication mode for medical sensor networks. Most existing broadcast protocols for wireless sensor networks use a minimal virtual backbone subnetwork to broadcast packets, which can minimize the total number of transmissions. However, since the unreliability of wireless link is not considered, the broadcast efficiency of these protocols is not high in factual networks. This paper proposes an efficient networkwide reliable broadcast protocol (ENWRB), which adopts network coding scheme to reduce the retransmissions in single hop broadcast and a link-quality-aware virtual backbone election algorithm to generate the more efficient broadcast backbone. Simulation results show that, under the same premise of ensuring all nodes receive broadcast packets successfully, the efficiency of ENWRB is higher than that of Hierarchical CDS-based Algorithm (HCA) greatly.
Authors: Xinguo Wang (Chengdu University of Information Technology), Run Hu (Chengdu University of Information Technology), Lutao Wang (Chengdu University of Information Technology), Dongrui Gao (Chengdu University of Information Technology), Yuyuan Su (Chengdu University of Information Technology), Bin Yang (Chengdu University of Information Technology),
Hide Authors & Abstract

Although anchors are the most common in cooperative localizations, they are not the optimal in the class of equality constraints which provide the global reference information for deriving absolute locations. Using Cram\'{e}r-Rao lower bound (CRLB) to evaluate the localization accuracy, this paper derives the optimal equality constraints that achieve the lowest CRLB trace under given constraint number, and analyzes the feasibility of constructing the optimal constraints before knowing the node ground truth locations. Simulations compare the performance between the anchor-type constraints and the optimal ones, and suggest a cooperative localization algorithm by using the optimal equality constraints.
Authors: ping zhang (Anhui Polytechnic University), Fei Cheng (Anhui Polytechnic University), Jian Lu (Southeast University),
Hide Authors & Abstract

. This paper focuses on the return link of a GEO satellite system that collects information from a large number of IoT devices that are sparsely distributed in a large geographical area. Narrow-Band (NB) IoT transmissions, with suitable Modulation and Coding Scheme (MCS), are Detected-and-Forwarded onboard the satellite, mapping each QAM symbol of the radio access link (uplink) into another equivalent PAM symbol that is suitable to modulate the intensity of the optical feeder link (downlink). Given the very large number of IoT devices that is expected to be served by the GEO satellite system, the feeder link (downlink) of the return channel is expected to be bottleneck. To tackle this limitation, apart from using an optical feeder link, time-packing is proposed to reduce the transmission time of the feeder link (downlink) without increasing the signal bandwidth and augment the number of IoT devices that could be simultaneously served in the radio access links (uplink). The Inter-Symbol Interference (ISI) that the time-packed feeder link generates is mitigated in part at the satellite gateway, using for this purpose an adaptive linear equalizer. After optical-to-electrical conversion, the NB-IoT codewords that are received in the gateway are decoded, correcting simultaneously errors introduced in both radio access and optical feeder links. Thus it is evaluated NBIoT's MCS to correct the errors using large overlapping factors to increase the optical feeder link data rate.
Authors: Joan Bas (CTTC), Alexis Dowhusko (Aalto University),
Hide Authors & Abstract

Decentralized Brains is a concept developed for multiple parallel control of decentralized collaborative swarms and systems. This systems communication paradigm comprises of local peer-to-peer control as well as the global state management which is required for large-scale collaborative systems. The scenarios vary from self-assembly protocols for aerospace structures to organizing a warehouse in a material-handling context where heterogeneous systems collaboratively accomplish a task. A reference implementation of the conceptualized protocol is developed and deployed in a 345 node testbed. A reliable broadcast communication primitive using synchronous broadcast is deployed in a dual-band SoC microcontroller. The performance of the adopted synchronous broadcast for network-wide flooding and consensus is presented in this article. The firmware is based on the latest branch of Contiki-NG to keep further open source implementations easier and modularized as per the ISO OSI networking model. Using the concepts of multi-hop mesh networking, network flooding using synchronous broadcasts from wireless sensor networks and multi-band radio controllers for cognitive radios, a hardwar software architecture is developed, deployed and evaluated. The synchronous broadcast has a success rate of more than 95% in network wide floods and the implicit network wide time synchronisation of less than 1~$\mu$s which is evaluated using experiments using a 345 node test bed is presented in this paper.
Authors: Aswin Karthik Ramachandran Venkatapathy (TU Dortmund University), Anas Gouda (TU Dortmund), Michael ten Hompel (Fraunhofer Institute for Material flow and Logistics), Joseph Paradiso (Media Lab),
Hide Authors & Abstract

Water is a natural resource necessary for life that must be taken care of. In coastal areas with near agricultural activity, it is very common to detect spills of chemical products that affect water quality of rivers and beaches. This water usually reaches the sea with bad consequences for nature and, there-fore, it is important to detect where possible spills are taking place and water does not have enough quality to be used. This paper presents the develop-ment of a LoRa (Long Range) based wireless sensor network to create an ob-servatory of water quality in coastal areas. This network consists of wireless nodes endowed with several sensors that allow measuring physical parame-ters of water quality, such as turbidity, temperature, etc. The data collected by the sensors will be sent to a gateway that will redirect them to a database. The database creates an observatory that will allow monitoring the environ-ment where the network is deployed in real time. Finally, the developed sys-tem will be tested in a real environment for its correct start-up. Two different tests will be performed. The first one will check the correct operation of sen-sors and network architecture; the second test will check the network cover-age of the commercial devices.
Authors: Sandra Sendra (Universidad de Granada), Marta Botella-Campos (Universitat Politecnica de Valencia, Spain), Jaime Lloret (Integrated Management Coastal Research Institute, Universidad Politecnica de Valencia, Spain), Jose Miguel Jimenez Herranz (Spain),
Hide Authors & Abstract

With the increasing usage of sensors and wireless communication systems, predictive maintenance is acquiring more and more importance to assess the condition of in-service equipments. Predictive maintenance presents promising cost savings, as it allows minimizing unscheduled systems or equipment failures, which can have very costly and catastrophic consequences in industrial environments. In parallel, with the advent of Industry 4.0, a tight integration of Internet of Things and Big Data analytics solutions, has become necessary to effectively manage industrial processes, and to early predict equipment faults or service disruptions. This paper presents a data-driven approach, based on multiple-instance learning, for predicting equipment failures by mining equipment event logs, which, while usually not designed for predicting failures, contains rich operational information. For evaluation, it was used a real-life dataset, with thousands of log messages, from an end-ofline testing equipment in a real automotive industry environment. The insights gained from mining such data will be shared on this paper, highlighting for the main challenges and benefits, and consequently, also good recommendations and practices for the appropriate usage of data analysis tools and techniques, will be shared.
Authors: David Vicêncio (Universidade de Trás-os-Montes e Alto Douro, Portugal), Hugo Silva (Continental Advanced Antenna, Portugal), Salviano Pinto Soares (Universidade de Trás-os-Montes e Alto Douro, Portugal), Vítor Filipe (Universidade de Trás-os-Montes e Alto Douro, Portugal), António Valente (Universidade de Trás-os-Montes e Alto Douro, Portugal),
Hide Authors & Abstract

State of the art internet of things (IoT) and mobile monitoring systems promise to help gathering real time progress information from construction sites. However, on remote sites the adaptation of those technologies is frequently difficult due to a lack of infrastructure and often harsh and dynamic environments. On the other hand, visual inspection by experts usually allows a quick assessment of a project’s state. In some fields, drones are already commonly used to capture aerial footage for the purpose of state estimation by domain experts. We propose a two-stage model for progress estimation leveraging images taken at the site. Stage 1 is dedicated to extract possible visual cues, like vehicles and resources. Stage 2 is trained to map the visual cues to specific project states. Compared to an end-to-end learning task, we intend to have an interpretable representation after the first stage (e.g. what objects are present, or later what are their relationships (spatial/semantic)). We evaluated possible methods for the pipeline in two use-case scenarios - (1) road and (2) wind turbine construction. We evaluated methods like YOLOv3-SPP for object detection, and compared various methods for image segmentation, like Encoder-Decoder, DeepLab V3, etc. For the progress state estimation a simple decision tree classifier was used in both scenarios. Finally, we tested progress estimation by a sentence classification network based on provided free-text image descriptions.
Authors: Peter Hevesi (German Research Center for Artificial Intelligence), Ramprasad Chinnaswamy Devaraj (German Research Center for Artificial Intelligence, TU Kaiserslautern), Matthias Tschöpe (German Research Center for Artificial Intelligence, TU Kaiserslautern), Oliver Petter (German Research Center for Artificial Intelligence, TU Kaiserslautern), Janis Elfert (German Research Center for Artificial Intelligence, TU Kaiserslautern), Vitor Fortes Rey (German Research Center for Artificial Intelligence), Marco Hirsch (German Research Center for Artificial Intelligence), Paul Lukowicz (German Research Center for Artificial Intelligence, TU Kaiserslautern),
Hide Authors & Abstract

In order to provide interactive capabilities within the context of Internet of Thing (IoT) applications, wireless communication systems play a key role, owing to in-herent mobility, ubiquity and ease of deployment. However, in order to comply with Quality of Service (QoS) and Quality of Experience (QoE) metrics, cover-age/capacity analysis must be performed, in order to account for the impact of signal blockage as well as multiple interference sources. This analysis is especial-ly complex in the case of indoor scenarios, such as those derived from Industrial Internet of Things (IIoT). In this work, a fully volumetric approach is employed in order to provide precise wireless channel characterization and hence, system level analysis of indoor scenarios. The proposed methodology will be tested against a real measurement scenario, providing full flexibility and scalability for adoption in a wide range of IIoT capable environments.
Authors: Francisco Falcone (UPNA), Imanol Picallo (UPNA), Peio López Iturri (UPNA), Mikel Celaya-Echarri (Tecnologico de Monterrey), Leyre Azpilicueta (Tecnologico de Monterrey),
Hide Authors & Abstract

Due to the ability of 3D-printers to build a wide range of objects at low costs, many industries are rapidly adopting additive manufacturing. However due to their sensing and communications capabilities, 3D-printers are Internet of Things (IoT) devices that are vulnerable to sophisticated cyberattacks, such as defect injection attacks. By maliciously manipulating the behavior of a 3D-printer, an attacker can compromise the integrity of a manufactured objects. To avoid detection, the adversary also compromises the sensor data reported by the 3D-printer that the operator could use to detect the attack. In this paper, we design a deep neural network that can detect such attacks by predicting the power consumption of a 3D-printer based on the object design and previous power consumption observations. By analyzing the difference between the predicted power consumption and the observed one, we can determine if the 3D-printer is under attack. By measuring the power consumption of the 3D-printer at the power line with an independent sensor, we can determine the true behavior of the 3D-printer without relying on sensor data reported by the potentially compromised 3D-printer. Compared to previous works, our proposed detection technique only requires cheap power sensors that can be easily installed. We conduct extensive experiments on a real-world additive manufacturing testbed and observe that our proposed method can detect defect injection attacks with up to 96% accuracy.
Authors: Michael Rott (Wichita State University), Sergio Salinas Monroy (Wichita State University),
Hide Authors & Abstract

The growing trend in the world population causes an increment in the demand for food as fruits like oranges. In this context, crop grass coverage becomes essential to reduce the resources in tree maintenance and improve the harvest. In this paper, we propose the use of remote sensing for monitoring the grass coverage. To do so, we have compared in time between plots with and without initial coverage and plots. In our work, we present image-processing techniques that consist of using different bands of Sentinel-2 images for different periods of the year, of obtaining reliable information fo changes in the grass-covered of the selected plots. The pixels of the selected images have a resolution of 10mx10m wherein our experiment represents information about orange trees plus grass coverage or soil. In addition, we present the results of the study, demonstrating the best behaviour of the presented technique. For this experiment, we use five different brands of the satellite, using red band, green band, blue band, near-infrared band, and water vapour band as well as normalised vegetation index using combinations of red and infrared bands. The significance values are obtaining applying Single Analysis of Variance, a Statistical analysis. In this case, the higher results are located in WVP band with Reason-F of 42.56 and Value-P of 0.000 blue band with Reason-F of 38.61 and Value-P of 0.000.
Authors: Daniel Andoni Basterrechea (Universitat Politecnica de Valencia, Spain), Lorena Parra (Integrated Management Coastal Research Institute, Universidad Politecnica de Valencia, Spain), Mar Parra (Universitat Politecnica de Valencia, Spain), Jaime Lloret (Integrated Management Coastal Research Institute, Universidad Politecnica de Valencia, Spain),
Hide Authors & Abstract

In the urban areas, the use of water to irrigate the green areas must be improved by the use of technology to reach water efficiency. Normalized Difference Vegetation Index is the most important indexes to evaluate the vegetation vigour, but the required equipment for its gathering have a high cost. In this paper, we present the use of Normalized Difference Vegetation Index and pictures taken with a regular camera to evaluate the status of two groups of plots under different solar exposure. Besides, we study the possibilities to correlate data obtained from regular pictures with Normalized Difference Vegetation Index, offering a low-cost option for monitoring plant status. From the 18 evaluated plots, which include 3 different grass combinations, the mean value of Normalized Difference Vegetation Index and one picture is taken. Then, we obtain the red, green, and blue histograms of each picture using Matlab software. The histograms were included in Statgraphics to search for correlations between histograms and Normalized Difference Vegetation Index of each plot. The highest correlation was found with the data of red histogram (R2=0.58 and high significance level). Finally, the variance of both evaluated variables is analyzed, and we have determined that both variables are useful in determining the solar exposure of studied plots. Significance level was higher in NDVI than with data of the histogram, but both of them have a P-Value lower than 0.05 in the analysis of variance.
Authors: José F. Marín (Area verde MG Projects SL., Spain), Lorena Parra (Universitat Politecnica de Valencia, Spain), Jaime Lloret (Universitat Politècnica de València, Spain), Salima Yousfi (Instituto Madrileño de Investigación y Desarrollo Rural, Agrario y Alimentario, Spain), Pedro V. Mauri (Instituto Madrileño de Investigación y Desarrollo Rural, Agrario y Alimentario, Spain),
Hide Authors & Abstract

The use of LoRa sensors and IoT in farming is increasing progressively. In this work we installed a series of soil moisture and conductivity sensor at 5 cm and 30 cm depth in a Camelina sativa (L.) Crantz cultivar. The information gathered by the sensors show how rain or irrigation water infiltrates in the soil. This al-lows the farmer to take decisions regarding the use of water in a very effective, cheap and reliable way.
Authors: David Mostaza-Colado (IMIDRA, Spain), Pedro V. Mauri (IMIDRA, Spain), Aníbal Capuano (Camelina Company Spain),
Hide Authors & Abstract