MQTT over QUIC protocol has recently garnered attention in the IoT field. This protocol enables faster communication than the traditional MQTT. However, the security risks arising from the new combination of MQTT and QUIC still need to be explained. In particular, the MQTT over QUIC broker's availability could be compromised by initial attacks that exploit specific vulnerabilities of the QUIC protocol. Initial attacks target the handshake process of the QUIC protocol and unfairly consume server-side resources. This study examines the impact of initial attacks on MQTT over QUIC brokers and proposes defensive measures. Our verification shows that initial attacks can indeed compromise the availability of MQTT over QUIC brokers. Furthermore, we confirmed that maintaining session continuity can preserve broker availability against initial attacks.

In incineration plants, indoor positioning systems are needed to prevent workers from approaching hazardous areas and to facilitate rescue operations in emergencies. Although many positioning systems using signal strength indicators such as Wi-Fi Received Signal Strength Indicator (RSSI) have been proposed, the vast interior of incineration plants increases the cost of constructing a communication infrastructure using $\mathbf{5 G H z}$ Wi-Fi. Therefore, by adopting Low Power Wide Area (LPWA), which allows for long-distance communication, the number of Access points (APs) that need to be installed in the environment can be reduced, thus lowering the cost of constructing the communication infrastructure. IEEE 802.11ah (11ah) is a new LPWA standard that uses the Sub-GHz band. In this study, we deployed new 11ah receivers in an incineration plant where IEEE 802.11ac (11ac) is already established and evaluated their coverage and area classification performance. The experiment demonstrated that the entire incineration plant can be covered with fewer units compared to 11ac. Additionally, when a small number of 11 ah receivers were installed, the system identified individual rooms with higher accuracy than 11ac, suggesting that 11ah RSSI is a promising feature that can be obtained at low cost.

There are broad benefits to developing pedestrian spaces in terms of environment, culture, and economy. Given this context, accurately measuring pedestrian traffic is considered a critical indicator of sidewalk usage. Currently, the mainstream method for detecting pedestrians utilizes RGB camera footage installed along sidewalks. However, especially during nighttime and adverse weather conditions, insufficient lighting hampers detection accuracy. In contrast, Far-Infrared(FIR) imaging does not require a light source as it measures radiated heat. This study proposes a pedestrian detection and tracking model that integrates the strengths of both RGB and FIR cameras through image fusion processing. Specifically, pedestrians are detected from fused images using a pedestrian detector and then tracked using a tracking system to measure pedestrian counts. Additionally, a version of the pedestrian detector trained on the fused images through transfer learning is developed, and its detection results are compared with those from a non-transfer learning model. The experimental results demonstrate that using fused images for detection and tracking is effective in specific data scenarios, confirming the utility of the image fusion model under varied conditions.

Cyberattacks on In-Vehicle Networks (IVNs) are becoming the most urgent issue. The Controller Area Network (CAN), one of the IVNs, is a standard protocol for automotive networks. Many researchers have tackled the security issues of CAN, such as the vulnerability of Denial-of-Service (DoS) attacks and impersonation attacks. Though existing methods can prevent DoS attacks, they have problems in deployment cost, isolability of a compromised Electronic Control Unit (ECU), and traceability for the root cause of isolation. Thus, we tackle to prevent DoS attacks on CAN. To solve these problems of the existing methods, we propose an isolable and traceable CAN-bus kernel-level protection called IVNPROTECT. IVNPROTECT can be installed on an ECU, which has a wireless interface, just by the software updating because it is implemented in the CAN-bus kernel driver. We also confirm that our IVNPROTECT can mitigate two types of DoS attacks without distinguishing malicious/benign CAN identifie rs. After mitigating DoS attacks, IVNPROTECT isolates a compromised ECU with a security error state mechanism, which handles security errors in IVNPROTECT. And, we evaluate the traceability that an ECU with IVNPROTECT can report warning messages to the other ECUs on the bus even while being forced to send DoS attacks by an attacker. In addition, the overhead of IVNPROTECT is 9.049 $00B5s, so that IVNPROTECT can be installed on insecure ECUs with a slight side-effect.

The Controller area network (CAN) is the most extensively used in-vehicle network. It is set to enable communication between a number of electronic control units (ECU) that are widely found in most modern vehicles. CAN is the de facto in-vehicle network standard due to its error avoidance techniques and similar features, but it is vulnerable to various attacks. In this research, we propose a CAN bus intrusion detection system (IDS) based on convolutional neural networks (CNN). U-CAN is a segmentation model that is trained by monitoring CAN traffic data that are preprocessed using hamming distance and saliency detection algorithm. The model is trained and tested using publicly available datasets of raw and reverse-engineered CAN frames. With an$F_{1} {Score}$of 0.997, U-CAN can detect DoS, Fuzzy, spoofing gear, and spoofing RPM attacks of the publicly available raw CAN frames. The model trained on reverse-engineered CAN signals that contain plateau attacks also results in a true positive rate and false-positive rate of 0.971 and 0.998, respectively.

Recently, cyberattacks on Controller Area Network (CAN) which is one of the automotive networks are becoming a severe problem. CAN is a protocol for communicating among Electronic Control Units (ECUs) and it is a de-facto standard of automotive networks. Some security researchers point out several vulnerabilities in CAN such as unable to distinguish spoofing messages due to no authentication and no sender identification. To prevent a malicious message injection, at least we should identify the malicious senders by analyzing live messages. In previous work, a delay-time based method called Divider to identify the sender node has been proposed. However, Divider could not identify ECUs which have similar variations because Divider's measurement clock has coarse time-resolution. In addition, Divider cannot adapt a drift of delay-time caused by the temperature drift at the ambient buses. In this paper, we propose a super fine delay-time based sender identification method with Time-to-Digital Converter (TDC). The proposed method achieves an accuracy rate of 99.67% in the CAN bus prototype and 97.04% in a real-vehicle. Besides, in an environment of drifting temperature, the proposed method can achieve a mean accuracy of over 99%.

Indoor localization has been a hot area of research over the past two decades. Since its advent, it has been steadily utilizing the emerging technologies to improve accuracy, and machine learning has been at the heart of that. Machine learning has been increasingly used in fingerprint-based indoor localization to replace or emulate the radio map that is used to predict locations given a location signature. The prediction quality of a machine learning model primarily depends on how well the model was trained, which relies on the amount and quality of data used to train it. Data augmentation has been used to improve quality of the trained models by synthetically producing more training data, and several approaches were used in the literature that tackles the problem of lack of training data from different angles. In this paper, we propose DataLoc+, a data augmentation technique for room-level indoor localization that combines different approaches in a simple algorithm. We evaluate the technique by comparing it to the typical direct snapshot approach using data collected from a field experiment conducted in a hospital. Our evaluation shows that the model trained using the proposed technique achieves higher accuracy. We also show that the technique adapts to larger problems using a limited dataset while maintaining high accuracy.

Controller Area Network (CAN) is one of the in-vehicle network protocols that is used to communicate among Electronic Control Units (ECUs) and has been de-facto standard. CAN is simple and has several vulnerabilities such as unable to distinguish spoofing messages because it doesn't support any authentication or sender identification properties. In previous work, some voltage-based methods to identify the sender node have been proposed. The methods can identify ECUs with high accuracy. However, the accuracy of source identification depends on a feature that is extracted from a continuous function of voltage use sampling. In general, as the sampling rate increases, the accuracy of identification is improved. Though the amount of data used for the identification increases too. Hence, it is desired to create an Intrusion Detection System (IDS) that identifies ECUs using few sampling features as there is a limited computing resource in vehicles. In this paper, we propose a delay-time based sender identification method of ECUs. We confirm that the proposed method achieved a true positive rate of 96.7% in CAN bus prototype against spoofing attack from a compromised ECU, detecting spoofing attack from an unmonitored ECU with a true positive rate of 98.0% in real-vehicle.

In search and rescue (SAR) operation, the potential of Unmanned Aerial Vehicles (UAVs) gathers great attention. Existing studies have made various experiments to find victims by a UAVs with a single sensor, e.g., one of an optical camera, a thermal infrared camera, and a radio wave signal. However, the experimental environments are limited to show the performance of the sensor. Since there are various SAR missions, it is difficult to choose the best sensor for all environments. Then, to enhance the UAV performance, we need to consider multiple sensors to find a victim efficiently. In the paper, we investigate optical camera and thermal infrared camera for finding a victim helpfully. In the investigation, we observed the differences between their images by distance and brightness to find a human.

In recent years, some bus companies have raised revenue by reviewing the route plan using the number of passengers. The company has a system that can automatically counts the number of passengers on an ongoing basis, but it is too expensive for bus companies that really need to reconsider their route planning to introduce the system. In order to solve this problem and realize efficient operation, we propose a method to count passengers by using a drive recorder and sensors those are already equipped with buses. Drive recorders and various sensors will be obliged by the government to be set up by transport operators in the future. We constructed a model using Random Forest Regression with the position of the bus from the GPS module in the buses, the position of the bus stop used for operation management, and the number of passengers estimated from image processing using background subtraction method. As a result, the average correct answer rate is 84%.

In bus companies, it is important for the operation manager to grasp operation states of the vehicle from the viewpoint of safety management and improving the operation efficiency. Currently, for allowing the operation manager to grasp operation states of the vehicle, the driver records operation states by manually operating a recorder called `Digital-tachograph'. However, operating a digital tachograph is a heavy burden to the driver. In this research, in order to solve this problem and to realize efficient operation, we propose a method for automatic classification of operation states using sensor data obtained from the buses. We implemented a classifier using Random Forest with the sensor data. The correct answer rate was 0.9 or more in each condition unless it was irregular operation.