Takeshi Kohama

Quantitative evaluation of arousal level based on the analyses of microsaccade rates and pupil fluctuations

In this study, we proposed an objective estimation of decline of arousal level by analyzing microsaccade rate and pupil fluctuation while subjects were continuously gazing a fixation target. Previous studies show that the slow eye movements (SEMs) could be a candidate for an indicator of decline of arousal. However, it is not sufficient to evaluate transition of arousal states since SEMs appear just prior to sleep onset. To establish more objective assessment of arousal, we examined the effects of the transition of arousal on microsaccade rate and pupil fluctuation. The subjects were instructed to indicate by mouse clicks when they were aware of having slept. We have analyzed the eye movement and pupil fluctuation data in advance of the occurrence of SEMs which were detected just before the mouse clicks. In the results, longitudinal pupil diameter shrinking and gradual rise of microsaccade rate were observed prior to SEMs. These results suggest that the arousal level could be evaluated by monitoring eye movements and pupil fluctuations.

Attentional effects on gaze preference for salient loci in traffic scenes

Alerting drivers for self-regulation of attention might decrease crash risks attributable to absent-minded driving. However, no reliable method exists for monitoring driver attention. Therefore, we examined attentional effects on gaze preference for salient loci (GPS) in traffic scenes. In an active viewing (AV) condition requiring endogenous attention for traffic scene comprehension, participants identified appropriate speeds for driving in presented traffic scene images. In a passive viewing (PV) condition requiring no endogenous attention, participants passively viewed traffic scene images. GPS was quantified by the mean saliency value averaged across fixation locations. Results show that GPS was less during AV than during PV. Additionally, gaze dwell time on signboards was shorter for AV than for PV. These results suggest that, in the absence of endogenous attention for traffic scene comprehension, gaze tends to concentrate on irrelevant salient loci in a traffic environment. Therefore, increased GPS can indicate absent-minded driving. Practitioner Summary: The present study demonstrated that, without endogenous attention for traffic scene comprehension, gaze tends to concentrate on irrelevant salient loci in a traffic environment. This result suggests that increased gaze preference for salient loci indicates absent-minded driving, which is otherwise difficult to detect.

Quantitative assessments of arousal by analyzing microsaccade rates and pupil fluctuations prior to slow eye movements

It is well known that a decline of arousal level causes of poor performance of movements or judgments. Our previous study indicates that microsaccade (MS) rates and pupil fluctuations change before slow eye movements (SEMs) (Honda et al. 2013). However, SEM detection of this study was obscure and insufficient. In this study, we propose a new SEM detection method and analyze MS rates and pupil fluctuations while subjects maintain their gaze on a target. We modified Shin et al.s method, which is optimized for EOG (electrooculography) signals, to extract the period of sustaining SEMs using a general eye tracker. After SEM detection, we analyzed MS rates and pupil fluctuations prior to the initiation of SEMs. As a result, we were able to detect SEMs more precisely than in our previous study. Moreover, the results of eye movements and pupil fluctuations analyses show that gradual rise of MS rate and longitudinal miosis are observed prior to the initiation of SEMs, which is consistent with our previous study. These findings suggest that monitoring eye movements and pupil fluctuations may evaluate the arousal level more precisely. Further, we found that these tendencies become more significant when they are restricted to the initial SEMs.

Extraction of relations between behaviors by lecturer and students in lectures

In this paper, we discuss the extraction of relations between lecturer and students in lectures by using multi-layered neural networks. Here, the relations among a few features concerning on the behaviors of the lecturer and students can be represented by multi-layered neural networks with the time-delay. Furthermore, we introduce a structural learning algorithm with forgetting for neural networks for the extraction of rules in the interaction. The above time-series models are analyzed focusing on weights in multi-layered neural networks. Concretely, we analyze which lecturers behaviors (face and hand movements and loudness of speech) give great influences on behaviors by students (face movements). On the contrary, we analyze which students behavior (face movement) give great influences on behaviors by lecturer.

Analysis of Non-verbal Behaviors by Students in Cooperative Learning

In this paper, we discuss the relationship between non-verbal behaviors and understandings by students in the cooperative learning. First, we detect non-verbal behaviors by students by using image processing methods. Next, we propose a modeling method for non-verbal behaviors. Furthermore, we discuss the relationship between non-verbal behaviors and understandings by students based on the above models.

Paper analysis of reading and writing behaviors for digital contents

In this paper, we discuss the change of page number and image processing methods for discrimination between behaviors in reading contents and taking notes for contents on the display. First, scene images for contents can be recorded by a sunglasses-type camera. Next, a few features can be extracted based on color information from these images and the discrimination procedure can be executed by the discrimination analysis and the classification method by using multi-layered neural networks. Finally, we evaluate the discrimination precision and discuss the relations among page transition, behaviors by students and their notes. This electronic document is a “live” template. The various components of your paper (title, text, heads, etc.) are already defined on the style sheet, as illustrated by the portions given in this document.

Analysis of behaviors by lecturer and students in lectures based on piecewise auto-regressive modeling

In this paper, we discuss the analysis of behaviors by lecturer and students in lectures. In lectures, the relationships between lecturer and students depend on not only contents but also the loudness of speech and gestures by lecturer. The behaviors consisting of head and eye movements by students can often represent how they have the interest with the lecture. First, we extract the loudness of speech and the face direction as behaviors by lecturer and the face direction as behaviors by students. Next, we construct piecewise auto-regressive (AR) models for their behaviors. Here, the piecewise AR models can be constructed based on the residual error. Finally, we show the analysis results for a real lecture based on the piecewise AR modeling.

Evaluating activities of daily living using an infrared depth sensor: KINECTTM

Abstract Purpose: With a growing proportion of elderly people in the population, the maintenance of activities of daily living (ADLs) in elderly people is crucial to keep medical costs down. We investigated the ADL measurement accuracy of KINECTTM and Kinect Studio. To eliminate the subjectivity of conventional methods, we numerically assessed motions with computer analysis. Methods: Eighteen actions that repeated “move” and “stationary” phases, including movement of arms, legs, head and torso were measured using KINECTTM. Errors and standard deviations of joint coordinates at the stationary points outputted from KINECT Studio were evaluated. Simultaneous measurements were performed with KINECTTM using conventional high-performance motion capture, and the output was treated as a true value for comparison. Results: In most motions, errors of the joint coordinates were within 100 mm; however, there were two cases where errors due to the skeleton-model estimation by KINECT Studio increased. Firstly, when a part of the body unexpectedly moved out of the infrared measurable area, and secondly, when parts of the body overlapped each other on the KINECTTM image. Conclusions: KINECTTM and Kinect Studio are effective for ADL assessment when positions that cause large errors are excluded. Since KINECTTM has sufficient precision, it should also be possible to develop a more appropriate ADL evaluation system with a new algorithm of skeleton-model estimation that does not depend on KINECT Studio. Implications for Rehabilitation The KINECTTM and Kinect Studio are effective for ADL assessment when positions that cause large errors are excluded With an increasing proportion of elderly people in the population, the maintenance of activities of daily living (ADLs) in elderly people is crucial to keep medical costs down Systems such as the KINECTTM can support these goals

Analysis of Behaviors of Participants in Meetings

In this paper, we analyze the behaviors of participants in two types of meetings (brainstorming and decision-making). First, we introduce the use of participant behavior based on facial movement. Next, we propose a method for modeling the behaviors of participants based on multi-layered neural networks. Lastly, based on our experimental results, we discuss the relationships between the meeting phase, participant behaviors, and the model parameters in these two types of meetings. Our results show the parameters in the above models to be strongly related to the behaviors and ideas of the participants in these two types of meetings.

Relationships between behaviors by applicants and interviewer in interview

In interviews, the evaluation by the interviewer is affected by not only the reply but also behaviors (eye and face movements) by applicants. Moreover, the evaluation by the interviewer is communicated with applicant as the behavior by the interviewer. Therefore, the interaction between the behaviors by the interviewer and the applicant occurs. In this paper, we show that the evaluation by the interviewer is strongly influenced by behaviors by applicants and the relationship between the evaluation by the interviewer and the behaviors by applicants.