Search

The conventional positioning methods, such as fingerprinting and trilateration, are commonly used in existing Wi-Fi positioning systems. Although the fingerprinting method offers relatively high accuracy, it faces challenges due to its sensitivity to environmental changes and the necessity of extensive training data and calibration. The trilateration method calculates positions based on the distances between anchors and targets. However, inaccuracies in measuring these distances could significantly impact the overall accuracy. Additionally, the necessity for at least three anchors creates a requirement for a more extensive infrastructure, posing challenges to practical service deployment. In this paper, we introduce bilatangulation, a novel cluster-based double-step positioning method that leverages distances calculated using fine timing measurement (FTM) and angles determined using channel state information (CSI) from two anchors. The first step addresses the symmetry problem of the two intersections in distance-based positioning by utilizing the angle orientations. In the second step, we performed measurement error compensation by clustering multiple intersections generated from both distance and angle data, taking into account the characteristics of each cluster. Our practical experiment was conducted indoors using off-the-shelf network interface card (NIC). For positioning, only two anchors were employed, resulting in an original mean positioning error (MPE) of $1.58\ m$ . Applying a measurement error compensation step reduced the final MPE by 88% compared to the original MPE.

Machine learning’s exponential growth has transformed healthcare, with Federated Learning (FL) playing a pivotal role. Despite its significance, FL is vulnerable to privacy attacks. In response, researchers have integrated differential privacy (DP) into FL. Nevertheless, incorporating DP introduces challenges such as increased total communication costs and computational overheads due to the introduction of noise. This drawback renders FL with DP less viable for healthcare systems, characterized by numerous low-resource devices and network bandwidth constraints. To overcome this limitation, we propose integrating a Discrete Fourier Transform (DFT) aggregator post-noise addition to transform the gradient generated by local training before sending it to the central server. This process reduces the gradient size and provides rudimentary encryption. The evaluation results reveal the superior performance of our proposed method, demonstrating an enhanced accuracy ranging from 0.2% to 2% compared to existing differential privacy techniques, including RDP, DP-SGD, ZcDP, LDP-Fed, and DP-AdapClip. Our approach substantially reduces the total communication costs (ranging from 6% to 43% across different privacy budgets) with faster training times in healthcare datasets such as the PIMA Indian database and Breast Cancer Histopathology Images.

In modern transportation, autonomous driving technology has emerged as a significant innovation, offering improved safety, efficiency, and convenience. One of its core technologies is path planning, which creates an optimal route for vehicles in various traffic situations, ensuring they arrive safely and efficiently. However, existing path planning methods suffer from inflexible and inefficient road sampling, and limitations in path expression due to curve models. In this paper, we introduce a novel path planning framework that operates from a coarse-to-optimal approach. We first introduce an adaptive sampling method to construct a three-dimensional (3D) space-aware profiling map in the Frenet frame to quantify abstract traffic scenarios. This cost map guides the generation of waypoints by incorporating obstacle distributions in the $s$ -direction and cost values in the $l$ -direction. We then design a cost function to obtain the coarse path by considering ride comfort, anti-deviation from the center of the road, and path safety. Finally, a new path model is re-established based on the Taylor series, and a quadratic programming approach is employed to fine-tune the coarse path to attain the optimal path. Experimental results show that the proposed adaptive sampling method can dynamically adjust the sampling strategy according to the change in the number and distribution of obstacles on the road, significantly improving the sampling effectiveness rate to 100%, 99.85%, 99.68% and 99.61% for one-obstacle, two-obstacle, three-obstacle, and four-obstacle scenarios. Compared with the conventional method, the generated path length is shortened by 3.48% and 8.42%, and the maximum heading angle is reduced by 18.68% and 30.21% in the scenarios with obstacles of the same and different sizes.

Subtitles play a crucial role in facilitating the understanding of visual content when watching films and television programs. In this study, we propose a method for presenting subtitles in a way that considers cognitive load when viewing video content in a non-native language. Subtitles are generally displayed at the bottom of the screen, which causes frequent eye focus switching between subtitles and video, increasing the cognitive load. In our proposed method, we focused on the position, display time, and amount of information contained in the subtitles to reduce the cognitive load and to avoid disturbing the viewer’s concentration. We conducted two experiments to investigate the effects of our proposed subtitle method on gaze distribution, comprehension, and cognitive load during English-language video viewing. Twelve non-native English-speaking subjects participated in the first experiment. The results show that participants’ gazes were more focused around the center of the screen when using our proposed subtitles compared to regular subtitles. Comprehension levels recorded using LightSub were similar, but slightly inferior to those recorded using regular subtitles. However, it was confirmed that most of the participants were viewing the video with a higher cognitive load using the proposed subtitle method. In the second experiment, we investigated subtitles considering connected speech form in English with 18 non-native English speakers. The results revealed that the proposed method, considering connected speech form, demonstrated an improvement in cognitive load during video viewing but it remained higher than that of regular subtitles.

Quadriplegics face a communication obstacle as their physical abilities are restricted, leaving them unable to speak or use their limbs, with only their upper neck being mobile. So, we propose a recognition system and a new communication language utilizing Morse code and head movements, to break this barrier. We aim to overcome the limitations of camera-based and wearable-sensor methods, including occlusion, privacy concerns, and user inconvenience. The goal is to passively detect quadriplegics’ head movements and map them to their corresponding character. The dataset including all 26 alphabet letters, was gathered in various settings, including single-user and multi-human environments, with multiple locations for each setting. For evaluation, 2% samples are randomly selected from the unseen environment to be used with the seen environment as a training dataset. Based on the results, our system demonstrates practical feasibility for real-world implementation, with accuracy rates of 94% and 80% achieved in single-user and multi-human environments, respectively.

Autonomous vehicles are an essential tool for promoting the development of intelligent transportation systems (ITS) and can effectively reduce traffic accidents caused by human errors. As an important part of the automatic driving software system, path planning is responsible for generating the motion trajectory of the vehicle, which is the primary factor determining driving quality. However, solution space construction and optimization problem formulation remain challenging research areas in the field of path planning. In this paper, we propose a multi-objective optimization algorithm for static obstacle avoidance to improve the comfort, safety and anti-deviation of the planned trajectory. We decouple the lateral and longitudinal motion of the vehicle using the Frenet frame and discretize the driving state space to generate target states of the vehicle. Based on the initial and target states, we generate a set of lateral and longitudinal motion trajectories using quintic and quartic polynomials, respectively. In addition, we design a cost function that comprehensively considers the comfort, safety, and deviation distance of the road center line by combining an acceleration check, curvature check, and collision check. As part of the cost function, we propose a novel method to quantify the safety of candidate trajectories considering the size of obstacles. The experimental results show that the proposed algorithm can quantize the safety of candidate paths and improve comfort 13.47%, 32.19%, 59.36% and 18.60% on a straight road, curvy road, intersection and U-shaped road, respectively. Furthermore, the algorithm can improve anti-deviation by 63.72%, 13.86%, 44.36%, and 45.56% on a straight road, curvy road, intersection and U-shaped road, respectively.

Learning user attributes is essential for providing users with a service. In particular, for e-commerce portals which deal in variety of goods ranging from clothes to foods to home electronics, it is especially important to learn “domain-independent” attributes such as age, gender, and personality that affect people’s behavior across various domains of daily life (e.g., clothing, eating and housing) because these attributes can be used for personalization in diverse domains their service covers. Thus far, researchers have proposed approaches to learn user representation (UR) from user-item interactions, trying to embed rich information about user attributes in UR. However, very few can learn URs that are domain-independent without confounding them with domain-specific attributes (e.g., food preferences). This could consequently undermine the former’s utility for personalizing services in other domains from which the URs are not learned. To address this, we propose an approach to learn URs that exclusively reflect domain-independent attributes. Our approach introduces a novel multi-layer RNN with two types of layers: Domain Specific Layers (DSLs) for modeling behavior in individual domains and a Domain Independent Layer (DIL) for modeling attributes that affect behavior across multiple domains. By exchanging hidden states between these layers, the RNNs implement the process of domain-independent attributes affecting domain-specific behavior and makes the DIL learn URs that capture domain-independence. Our evaluation results confirmed that the URs learned by our approach have greater utility in predicting behavior in the other domains from which these URs were not learned thereby demonstrating adaptability to various domains.

It is necessary to optimize evacuation guidance to shelters in short evacuation time. The state-of-the-art method based on an idea of combinatorial optimization problems related to evacuees’ locations and the capacities of nearby shelters has been developed, while it cannot mitigate the effect of congestion on roads/streets after evacuation starts. In this study, to cover this problem, we develop a new method that utilizes simulations for estimating the effect of congestion on roads/streets during evacuation and reassigning shelters to evacuees based on the simulation results. By iterating this step, our method derives the congestion-aware solutions for shelter selection that can realize more smooth evacuation. To evaluate our method, we conducted multi-agent simulations assuming a disaster situation in a sightseeing spot. Specifically, we examined a hypothetical case scenario involving the evacuation of 30,000 visitors from the Gion Festival. We compared the proposed method with conventional methods, such as the nearest shelter selection method and our previous method. We found that our proposed method reduced average and total evacuation time and congestion on roads compared to the conventional methods including the nearest shelter selection method and our previous method that only employs combinatorial optimization without estimating congestion. From this result, our idea of simulation-based congestion estimation has an impact of easing congestion during evacuation and preventing overcapacity of shelters at the same time. It shows the possibilities of help in developing congestion-aware evacuation strategies in emergency situations of crowded areas like huge cities or sightseeing spots.

The analysis of the movement of people in a shopping area with the aim of improving marketing is an important research topic. Many conventional methods are dependent on the density of people in the area, which is easily estimated by counting the people entering or exiting the area. However, a high density does not always mean an increase in activity, as certain people are simply passing the area at a given time. The primary goal of this study was to introduce a set of indicators for measuring the bustle of the area, which we call “Nigiwai,” from pedestrian movement by using an analogy from classical kinematics. Such indicators can be used to measure the impact of promotional events and to optimize the design of the area. Our novel indicators were evaluated with simulated pedestrian scenarios and were demonstrated to distinguish shopping scenarios from those in which people move around without shopping successfully, even when the latter scenarios had much higher densities. The indicators were computed solely from the pedestrian trajectory, which can easily be obtained from ordinary sensors using deep learning-based techniques. As a demonstration with real data, we applied our method to a video of a street and provided a visualization of the indicators.

Numerous technologies exist for promoting a healthier lifestyle. These technologies collectively referred to as “Behavior Change Support Systems”. However, the majority of existing apps use quantitative data representation. Since it is difficult to understand the meaning behind quantitative data, this approach has been suggested to lower users’ motivation and fail to promote behavior change. Therefore, an interpretation of quantitative data needs to be provided as a supplement. However, different descriptions of the same data may lead to different outcomes. In this paper, we explore the impact of different communication styles for interpretations of quantitative data on behavior change by developing and evaluating Walkeeper – a web-based app that provides interpretations of the users’ daily step counts using different levels of elaborateness and indirectness with the aim of promoting walking. Through the quantitative analysis and results of a user study, we contribute new knowledge on designing such interpretations for quantitative data.

As society grows ever more interconnected, the need for sophisticated signal processing and data analysis techniques becomes increasingly apparent. This is particularly true in the field of intelligent transportation systems (ITSs), where various sensing applications generate data at an exponential rate. In this paper, we present C-AVDI, a compressive measurement-based acoustic vehicle detection and identification architecture capable of extracting information from vehicle audio signals while sampling at sub-Nyquist rates. In addition, we further reduce the overall complexity by performing any necessary signal filtering during the acquisition process, removing the need for a separate filtering stage in the system’s front-end. Our results obtained from data collected under a range of weather conditions present an accuracy of 80% with a back-end analog-to-digital converter (ADC) sample rate of 3kHz, with initial results from a microcontroller (MCU) implementation of our proposed system presenting an accuracy of 72%.

Some respondents make careless responses due to the “satisficing,” which is an attempt to complete a questionnaire as quickly and easily as possible. To obtain results that reflect a fact, detecting satisficing and excluding the responses with satisficing from the analysis targets are required. One of the devised methods detects satisficing by adding questions that check violations of instructions and inconsistencies. However, this approach may cause respondents to lose their motivation and prompt them to satisficing. Additionally, a deep learning model that automatically answers these questions was reported. This threatens the reliability of the conventional method. To detect careless responses without inserting such screening questions, machine learning (ML) detection using data obtained from answer results was attempted in a previous study, with a detection rate of 55.6%, which is not sufficient from the viewpoint of practicality. Therefore, we hypothesized that a supervised ML model with a higher detection rate could be constructed by using on-screen answering behavior as features. However, (1) no existing questionnaire system can record on-screen answering behavior and (2) even if the answering behavior can be recorded, it is unclear which answering behavior features are associated with satisficing. We developed an answering behavior recording plug-in for LimeSurvey, an online questionnaire system used all over the world, and collected a large amount of data (from 5,692 people) in Japan. Then, a variety of features were examined and generated from answering behavior, and we constructed ML models to detect careless responses. We call this detection method the ML-ABS (ML-based answering behavior scale). Evaluation by cross-validation demonstrated that the detection rate for careless responses was 85.9%, which is much higher than the previous ML method. Among the various features we proposed, we found that reselecting the Likert scale and scrolling particularly contributed to the detection of careless responses.

In this paper, aiming to support easy walking route planning, we propose methods for predicting a heart rate along the arbitrary route without walking data, and recommending a semi-optimal walking route based on the predicted results. In our method, we build a model to predict the heart rate during walking with expected walking speed and gradient along a target route, and compute a semi-optimal walking route (near least physical load route satisfying calorie/distance constraints requested by a user) by using the model. In order to evaluate the accuracy of the prediction model, a walking experiment with 39 participants was conducted. The result showed that our model could predict the heart rate with mean absolute error (MAE) of 6.31 beats per minute on average. We also confirmed that the route recommended by our method satisfied calorie/distance constraints requested by a user while keeping the average and the maximum physical load (in terms of heart rate reserve) at 29.5% (light load) and 44.4% (moderate load), respectively.

Social support can improve adherence to physical activity and healthy eating behavior change, and recently, smartphone fitness applications which provide interventions to groups of users are increasingly popular. However, the types of social support provided in the applications, and the features/approaches used to offer the support, vary widely. It is unknown which approach is the best to meet users' requirements. We conducted a mixed methods study to identify user requirements for social support in group fitness applications. We then developed MyFitnessTeam based on these requirements and evaluated it in a six-week field study. We found that participants desired informational and emotional support, and for group fitness applications to provide daily challenges that they could complete with their support network. MyFitnessTeam significantly improved adherence to fitness behaviors, with men having significantly higher adherence levels than women and participants who were inactive at baseline achieving similar adherence levels to the most baseline-active participants. Groups of friends were more likely to exchange informational support but emotional support was low for both friends and strangers, due to lack of knowledge on when and how to offer such support. We propose five design recommendations to promote supportive interactions in group fitness applications.

The emergence of various types of commercial cameras (compact, high resolution, high angle of view, high speed, and high dynamic range, etc.) has contributed significantly to the understanding of human activities. By taking advantage of the characteristic of a high angle of view, this paper demonstrates a system that recognizes micro-behaviors and a small group discussion with a single 360 degree camera towards quantified meeting analysis. We propose a method that recognizes speaking and nodding, which have often been overlooked in existing research, from a video stream of face images and a random forest classifier. The proposed approach was evaluated on our three datasets. In order to create the first and the second datasets, we asked participants to meet physically: 16 sets of five minutes data from 21 unique participants and seven sets of 10 min meeting data from 12 unique participants. The experimental results showed that our approach could detect speaking and nodding with a macro average f1-score of 67.9% in a 10-fold random split cross-validation and a macro average f1-score of 62.5% in a leave-one-participant-out cross-validation. By considering the increased demand for an online meeting due to the COVID-19 pandemic, we also record faces on a screen that are captured by web cameras as the third dataset and discussed the potential and challenges of applying our ideas to virtual video conferences.

World health statistics about overweight and obesity show that overweight, obesity and diet-related diseases still remain major health risks. According to the World Health Organization, most of the world's population live in countries where overweight and obesity kill more people than underweight. Recently, many studies have been driven by the motivation of elaborating the “ideal” solutions to prevent and/or monitor overweight, obesity and diet-related diseases, and to encourage healthy diet and lifestyle. In this work, we present our idea to automatically measure food weight and calories, from food photo using ordinary chopsticks as a measurement reference. The analysis of the obtained results show that the use of near-by eating utensils combined with computer vision techniques is a great and exploitable approach to ubiquitously help in diet assessment and obesity treatment.

These days, smart home applications such as concierge service for residents, home appliance control, and so on are attracting attention. To realize these applications, we need a system which recognizes various human activities accurately with a low cost device. There are many studies on the activity recognition in a smart home. We also have proposed an activity recognition technique in a smart home by utilizing digital-output-PIR (passive infrared) sensors, door sensors, and power meters. However, the study has an unsolved issue: we cannot distinguish similar activities happening at the same place, for example, “eating” and “reading” while sitting on a sofa. In order to cope with this challenge, we introduce ALPAS: analog-output-PIR-sensor-based activity recognition technique which recognizes the different activities in the same place. Our technique recognizes user's activity by utilizing machine learning with frequency components of the sensor's output as features. However, because the number of features used in ALPAS is 1000 for each analog PIR-sensor, a large capacity memory is required. To reduce the number of features, we select a part of the sensing data. We call the starting point of the selected data as starting frequency (SF) and ending part as ending frequency (EF). We searched SF and EF using a grid search, and evaluated the recognition accuracy. We evaluated the proposed technique in a smarthome testbed. In the evaluation, five participants performed four different activities while sitting on a sofa. As a result, we achieved F-Measure: 63.9% when the EF is 1.4Hz, and F-Measure: 50% or lower when the SF is 9.9Hz or higher.

Changing behavior related to improper lifestyle habits has attracted attention as a solution to prevent lifestyle diseases, such as diabetes, heart disease, arteriosclerosis, and stroke. To drive health behavior changes, wearable devices are needed, and they must not only provide accurate sensing and visualization functions but also effective intervention functions. In this paper, we propose a health support system, WaistonBelt X, that consists of a belt-type wearable device with sensing and intervention functions and a smartphone application. WaistonBelt X can automatically measure a waistline with a magnetometer that detects the movements of a blade installed in the buckle, and monitor the basic activities of daily living with inertial sensors. Furthermore, WaistonBelt X intervenes with the user to correct lifestyle habits by using a built-in vibrator. Through evaluation experiments, we confirmed that our proposed device achieves measurement of the circumference on the belt position (mean absolute error of 0.93 cm) and basic activity recognition (F1 score of 0.95) with high accuracy. In addition, we confirmed that the intervention via belt vibration effectively improves the sitting posture of the user.

With the spread of smart devices, people may obtain a variety of information on their surrounding environment thanks to sensing technologies. To design more context-aware systems, psychological user context (e.g., emotional status) is a substantial factor for providing useful information in an appropriate timing. As a typical use case that has a high demand for context awareness but is not tackled widely yet, we focus on the tourism domain. In this study, we aim to estimate the emotional status and satisfaction level of tourists during sightseeing by using unconscious and natural tourist actions. As tourist actions, behavioral cues (eye and head/body movement) and audiovisual data (facial/vocal expressions) were collected during sightseeing using an eye-gaze tracker, physical-activity sensors, and a smartphone. Then, we derived high-level features, e.g., head tilt and footsteps, from behavioral cues. We also used existing databases of emotionally rich interactions to train emotion-recognition models and apply them in a cross-corpus fashion to generate emotional-state prediction for the audiovisual data. Finally, the features from several modalities are fused to estimate the emotion of tourists during sightseeing. To evaluate our system, we conducted experiments with 22 tourists in two different touristic areas located in Germany and Japan. As a result, we confirmed the feasibility of estimating both the emotional status and satisfaction level of tourists. In addition, we found that effective features used for emotion and satisfaction estimation are different among tourists with different cultural backgrounds.

This paper proposes a new replica placement algorithm that expands the exhaustive search limit with reasonable calculation time. It combines a new type of parallel data-flow processor with an architecture tuned for fast calculation. The replica placement problem is to find a replica-server set satisfying service constraints in a content delivery network (CDN). It is derived from the set cover problem which is known to be NP-hard. It is impractical to use exhaustive search to obtain optimal replica placement in large-scale networks, because calculation time increases with the number of combinations. To reduce calculation time, heuristic algorithms have been proposed, but it is known that no heuristic algorithm is assured of finding the optimal solution. The proposed algorithm suits parallel processing and pipeline execution and is implemented on DAPDNA-2, a dynamically reconfigurable processor. Experiments show that the proposed algorithm expands the exhaustive search limit by the factor of 18.8 compared to the conventional algorithm search limit running on a Neumann-type processor.