Research Results
Contents
- Biomarker detection for psychiatric disorder patients from invasive/non-invasive EEG data
Responsive deep brain stimulation guided by ventral striatal electrophysiology of obsession durably ameliorates compulsion
Contributions:
We reported a first-in-human application of responsive deep brain stimulation for OCD
The area-based feature triggered closed-loop stimulation to ameliorate OCD symptoms
We identified an association between NAc-VeP low-frequency oscillatory power and OCD
We provided a concept for a personalized, physiologically guided DBS strategy for OCD
Accumbens connectivity during deep-brain stimulation differentiates loss of control from physiologic behavioral states
Contributions:
Closed-loop therapies for binge-eating must predict onset of loss-of-control eating.
High-frequency phase-locking value in the nucleus accumbens predicts LOC eating.
High-frequency PLV is specific for LOC, distinguishing it from confounding states.
- Segmentation & Detection from real-world environments
Sequential thermal image-based adult and baby detection robust to thermal residual heat marks
Contributions:
We propose a sequential thermal image-based adult and baby detection method that outperforms existing object detection methods in terms of reducing the detection error due to thermal residual heat marks.
We verified the effectiveness of utilizing temperature differences for inducing robust detection to thermal residual heat marks.
Feed-forward hysteresis compensation using GAN-based surgical tool segmentation for surgical manipulation
Contributions:
A new pipeline is proposed for the hysteresis compensation of cable-driven flexible robots that can enhance their surgical manipulation abilities practically without any prior knowledge of the present robot shape or the use of external sensors
We propose a novel GAN-based surgical tool segmentation for hysteresis compensation.
A novel approach to surgical tools is proposed for efficient model learning with a novel training loss.
- Anomaly detection from human physiological and kinematical signals (ph.D dissertation)
GAN-based anomaly detection
Contributions:
This study primarily evaluates suitable GAN-based anomaly detection models for fall detection from among the nine recently proposed models.
This study proposes novel GAN-based anomaly detection model with user initial information (UI-GAN), which exhibits better results from the nine GAN-based anomaly detection models.
The merit of employing the UI heart rate information is demonstrated.
UI-GAN exhibits a better performance compared to eight recently proposed fall detection approaches.
(Ph.D Thesis) UI-GAN: Generative adversarial network based anomaly detection using user initial information for wearable devices
Cluster-analysis-based anomaly detection using heart rate sensor and accelerometer
Contributions:
We propose the best feature subset of heart rate and acceleration signals for the purpose of demonstrating reliable performance and designing a low-complexity model.
We verify that combining a heart rate sensor with an accelerometer increases the effectiveness of detecting falls.
We show the effectiveness of the user-adaptive method when using both heart rate and acceleration signals that has been hardly covered in other papers.
We prove that the performance of the proposed is better than that of recent user-adaptive and user-independent approaches by comparing with 12 conventional approaches.
(a) HAF (the proposed feature) variation between before and after falling in subjects 5, 7, and 10.
(b) Single accelerometer feature variation between before and after falling in subjects 5, 7, and 10.
(a) Acceleration change curves during the process of falling.
(b) Different phases within a fall event derived from HAF (the proposed feature).
(c) Different phases within a fall event derived from single accelerometer feature.
Learning-based fall detection using fusion of thermal imagery camera and accelerometer
Contributions:
We propose a fusion method using thermal image and accelerometer signal for fall detection.
We prove a higher detection accuracy when using fusion.
A patient's privacy is protected because we aim to use low resolution thermal imagery camera.
Fall detection using sensor fusion
Learning-based self-powered fall detection using pressure sensing
Contributions:
We firstly propose self-powered fall detection system which has no power source in the system.
Analytically selected decision boundary to distinguish falls and non-falls are proposed
Self-powered fall detection
Self-powered fall detection
Self-powered fall detection
Self-powered fall detection
- Thermal image-based person tracking and fall prediction
Learning-based person detection using thermal imagery camera
Contributions:
We propose learning-based automatic person detection using low resolution thermal imagery camera.
A patient's privacy is protected because we aim to use low resolution thermal imagery camera.
Person detection in a thermal image
Learning-based fall prediction using thermal imagery camera
Contributions:
We propose learning-based automatic fall prediction system using low resolution thermal imagery camera.
The proposed features can be efficiently detect whether stable or unstable status.
A patient's privacy is protected because we aim to use low resolution thermal imagery camera.
User image projection based fall prediction
- Touch gesture recognition
Touch gesture recognition in a different orientation settings
Contributions:
We propose a new touch gesture recognition system.
We note that touch sensors can be installed in various orientations in practical usages. We examine the effect of touch sensor orientation on recognition performance.
Touch gesture recognition using CNN
- Human-robot interaction
Research in robot emotion model for emotional interaction between human and robot
Contributions:
We propose a new emotion model for a mobile robot which makes well-communicated with human.
We demonstrate a user test for validate an effectiveness of our emotion model.
Emotion interaction between human and robot
Driving situation awareness-based human-robot interaction
Contributions:
Proper driving guidelines can be offered by the proposed system.
Hidden Markov models-based driving situation recognition has a reasonable accuracy.
We prove a real-time driving data such as an angle of steering wheel, speed, RPM can be effective material for the driver's driving situation.
Driving situation awareness using HMM for HRI