Drowsy driving is a major threat to the safety of drivers and road traffic. Accurate and reliable drowsy driving detection technology can reduce accidents caused by drowsy driving. In this study, we present a new method to detect drowsy driving with vehicle sensor data obtained from the steering wheel and pedal pressure. From our empirical study, we categorized drowsy driving into long-duration drowsy driving and short-duration drowsy driving. Furthermore, we propose an ensemble network model composed of convolution neural networks that can detect each type of drowsy driving. Each subnetwork is specialized to detect long- or short-duration drowsy driving using a fusion of features, obtained through time series analysis. To efficiently train the proposed network, we propose an imbalanced data-handling method that adjusts the ratio of normal driving data and drowsy driving data in the dataset by partially removing normal driving data. A dataset comprising 198.3 h of in-vehicle sensor data was acquired through a driving simulation that includes a variety of road environments such as urban environments and highways. The performance of the proposed model was evaluated with a dataset. This study achieved the detection of drowsy driving with an accuracy of up to 94.2%.

Air pollution is a social problem, because the harmful suspended materials can cause diseases and deaths to humans. Specifically, particulate matters (PM), a form of air pollution, can contribute to cardiovascular morbidity and lung diseases. Nowadays, humans are exposed to PM pollution everywhere because it occurs in both indoor and outdoor environments. To purify or ventilate polluted air, one need to accurately monitor the ambient air quality. Therefore, this study proposed a practical particulate matter sensing and accurate calibration system using low-cost commercial sensors. The proposed system basically uses noisy and inaccurate PM sensors to measure the ambient air pollution. This paper mainly deals with three types of error caused in the light scattering method: short-term noise, part-to-part variation, and temperature and humidity interferences. We propose a simple short-term noise reduction method to correct measurement errors, an auto-fitting calibration for part-to-part repeatability to pinpoint the baseline of the signal that affects the performance of the system, and a temperature and humidity compensation method. This paper also contains the experiment setup and performance evaluation to prove the superiority of the proposed methods. Based on the evaluation of the performance of the proposed system, part-to-part repeatability was less than 2 μg/m³ and the standard deviation was approximately 1.1 μg/m³ in the air. When the proposed approaches are used for other optical sensors, it can result in better performance.

Advances in vehicle technology have resulted in the development of vehicles equipped with sensors to acquire standardized information such as engine speed and vehicle speed from the in-vehicle controller area network (CAN) system. However, there are challenges in acquiring proprietary information from CAN frames, such as the brake pedal and steering wheel operation, which are essential for driver behavior analysis. Such information extraction requires electronic control unit identifier analysis and accompanying data interpretation. In this paper, we present a system for the automatic extraction of proprietary in-vehicle information using sensor data correlated with the desired information. First, the proposed system estimates the vehicle’s driving status through threshold-, random forest-, and long short-term memory-based techniques using inertial measurement unit and global positioning system values. Then, the system segments in-vehicle CAN frames using the estimation and evaluates each segment with our scoring method to select suitable candidates by examining the similarity between each candidate and its estimation through the suggested distance matching technique. We conduct comprehensive experiments of the proposed system using real vehicles in an urban environment. Performance evaluation shows that the estimation accuracy of the driving condition is 84.20%, and the extraction accuracy of the in-vehicle information is 82.31%, which implies that the presented approaches are quite feasible for automatic extraction of proprietary in-vehicle information.

As society has developed, the number of vehicles has increased and road conditions have become complicated, increasing the risk of crashes. Therefore, a service that provides safe vehicle control and various types of information to the driver is urgently needed. In this study, we designed and implemented a real-time traffic information system and a smart camera device for smart driver assistance systems. We selected a commercial device for the smart driver assistance systems, and applied a computer vision algorithm to perform image recognition. For application to the dynamic region of interest, dynamic frame skip methods were implemented to perform parallel processing in order to enable real-time operation. In addition, we designed and implemented a model to estimate congestion by analyzing traffic information. The performance of the proposed method was evaluated using images of a real road environment. We found that the processing time improved by 15.4 times when all the proposed methods were applied in the application. Further, we found experimentally that there was little or no change in the recognition accuracy when the proposed method was applied. Using the traffic congestion estimation model, we also found that the average error rate of the proposed model was 5.3%.

Dialogue systems must understand children’s utterance intentions by considering their unique linguistic characteristics, such as syntactic incompleteness, pronunciation inaccuracies, and creative expressions, to enable natural conversational engagement in child–robot interactions. Even state-of-the-art large language models (LLMs) for language understanding and contextual awareness cannot comprehend children’s intent as accurately as humans because of their distinctive features. An LLM-based dialogue system should acquire the manner by which humans understand children’s speech to enhance its intention reasoning performance in verbal interactions with children. To this end, we propose a fine-tuning methodology that utilizes the LLM–human judgment discrepancy and interactive response data. The former data represent cases in which the LLM and human judgments of the contextual appropriateness of a child’s answer to a robot’s question diverge. The latter data involve robot responses suitable for children’s utterance intentions, generated by the LLM. We developed a fine-tuned dialogue system using these datasets to achieve human-like interpretations of children’s utterances and to respond adaptively. Our system was evaluated through human assessment using the Robotic Social Attributes Scale (RoSAS) and Sensibleness and Specificity Average (SSA) metrics. Consequently, it supports the effective interpretation of children’s utterance intentions and enables natural verbal interactions, even in cases with syntactic incompleteness and mispronunciations.