DJDawoon Jung

Dawoon Jung

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Dawoon Jung has a total of 25 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Gait spatial parameter estimation from smart insoles using deep learning in a large heterogeneous clinical population

    by Ankhzaya Jamsrandorj, Dawoon Jung, Sungho Park, Beomjoon Park, Kyunghoon Kang, Jinwook Kim, Min Seok Baek, Kyung-Ryoul Mun

    While achieving reasonable accuracy, prior smart insole-based approaches for gait parameter estimation have been limited by small sample sizes, narrow clinical settings, and reliance on demographic or anthropometric priors. These limitations restrict their generalizability to heterogeneous real-world populations with diverse gait characteristics. Thus, this study proposes a smart insole-based gait parameter estimation approach that includes diverse health conditions while minimizing feature engineering and eliminating the need for demographic or anthropometric inputs. Using minimally engineered inertial measurement unit signals obtained from smart insoles worn by 702 individuals with diverse neurological and musculoskeletal conditions, a Conformer-based deep learning network was developed to estimate four spatial gait parameters: stride length, step length, stride width, and step width. The performance of the proposed approach was evaluated against the GAITRite walkway system and compared with previously suggested approaches, with additional evaluation using a publicly available dataset. The proposed approach showed excellent agreement with the reference system across walking speeds and evaluation levels. For example, subject-wise estimation of step length at the usual speed achieved a mean absolute error of 1.33 cm (ICC = 0.99). Scatter and Bland-Altman analyses further demonstrated strong agreement between the proposed approach and the reference system. These results highlight the potential of wearable smart insoles combined with deep learning for scalable gait parameter estimation.

  • Lower limb joint angle estimation using a single RGB camera

    by Vanyi Chao, Sungho Park, Ankhzaya Jamsrandorj, Dawoon Jung, Jinwook Kim, Kyung-Ryoul Mun

  • A vision-based system analyzing sit-to-stand and its clinical utility in classifying Parkinson’s disease

    by Yin May Oo, Ankhzaya Jamsrandorj, Heeeun Jung, Quynh Hoang Ngan Nguyen, Dawoon Jung, Dallah Yoo, Jinwook Kim, Tae-Beom Ahn, Kyung-Ryoul Mun


  • Open Access

    Smart insole-based abnormal gait identification: Deep sequential networks and feature ablation study

    by Beomjoon Park, Minhye Kim, Dawoon Jung, Jinwook Kim, Kyung-Ryoul Mun

    Objective: Gait analysis plays a pivotal role in evaluating walking abilities, with recent advancements in digital health stressing the importance of efficient data collection methods. This study aims to classify nine gait types including one normal and eight abnormal gaits, using sequential network-based models and diverse feature combinations obtained from insole sensors. Methods: The dataset was collected using insole sensors from subjects performing 15 m walking with designated gait types. The sensors incorporated pressure sensors and inertial measurement units (IMUs), along with the center of pressure engineered from the pressure readings. A number of deep learning architectures were evaluated for their ability to classify the gait types, focusing on feature sets including temporal parameters, statistical features of pressure signals, center of pressure data, and IMU data. Ablation studies were also conducted to assess the impact of combining features from different modalities. Results: Our results demonstrate that models incorporating IMU features outperform those using different combinations of modalities including individual feature sets, with the top-performing models achieving F1-scores of up to 90% in sample-wise classification and 92% in subject-wise classification. Additionally, an ablation study reveals the importance of considering diverse feature modalities, including temporal parameters, statistical features from pressure signals, center of pressure data, and IMU data, for comprehensive gait classification. Conclusion: Overall, this study successfully developed deep sequential models that effectively classify nine different gait types, with the ablation study underscoring the potential for integrating features from diverse domains to enhance clinical applications, such as intervention for gait-related disorders.


  • Open Access

    View-independent gait events detection using CNN-transformer hybrid network

    by Ankhzaya Jamsrandorj, Dawoon Jung, Konki Sravan Kumar, Muhammad Zeeshan Arshad, Hwasup Lim, Jinwook Kim, Kyung-Ryoul Mun


  • Open Access

    Deep Learning Networks for View-Independent Gait Events Detection

    by Ankhzaya Jamsrandorj, Dawoon Jung, Konki Sravan Kumar, Muhammad Zeeshan Arshad, Kyung-Ryoul Mun, Jinwook Kim

    Accurate gait detection is crucial in utilizing the ample health information embedded in it. As an alternative of the laborious and demanding sensor-based detection, vision-based approaches emerged.Yet, its complicate feature engineering process and heavy reliance on the lateral views serve as challenges. This study aimed to propose a view-independent vision-based gait event detection using deep learning networks that requires no pre-processing. A total of 22 participants performed seven different walking and running-related actions and the sequential video frames acquired from their actions were used as the input of the deep learning networks to produce the probability of gait events as outputs. The Transformer network and ResNet18 trained with sequential video frames achieved an F1-score of 0.956 or higher for walking straight and walking around. The detection performance on the frontal, lateral, and backside views did not differ much. The findings enhance applicability of vision-based approach and contribute to increasing its utility in health monitoring.


  • Personal Identification Using Gait Spectrograms and Deep Convolutional Neural Networks

    by Dawoon Jung, Mau Dung Nguyen, Muhammad Zeeshan Arshad, Jinwook Kim, Kyung-Ryoul Mun

    Human gait can serve as a useful behavioral trait for biometrics. Compared to fingerprint, face, and iris, the most commonly used physiological identifiers, gait can be unobtrusively monitored from a distance without requiring explicit involvement and physical restraint from people. Advances in wearable technology facilitate direct and faithful measurement of gait motions with easy-to-use and low-cost inertial sensors. This study aimed to propose an approach to using kinematic gait data collected with wearable inertial sensors for reliable personal identification. Sixty-nine individuals ranged in age from 24 to 62 years old participated in this study. The 3-axis acceleration and the 3-axis angular velocity signals were measured using the inertial measurement units attached to the feet, shanks, thighs, and posterior pelvis while walking. The gait spectrograms were acquired by applying time-frequency analyses to the lower body movement signals measured in one stride. Among each participant's 15 strides, 12 strides were used in the 4-fold cross validation of the deep convolutional neural network-based classifiers, and the remaining three strides were used to evaluate the classifiers. An accuracy of 99.69% was achieved by using the foot, shank, thigh, and pelvic spectrograms, and the accuracy was 96.89% using only the foot spectrograms. This study has the potential to be applied in behavior-based biometric technologies by confirming the feasibility of the proposed kinematic and spectrographic approaches in identifying personal behavioral characteristics.


  • Classifying the Risk of Cognitive Impairment Using Sequential Gait Characteristics and Long Short-Term Memory Networks

    by Dawoon Jung, Jinwook Kim, Miji Kim, Chang Won Won, Kyung-Ryoul Mun

    Cognitive impairment in the elderly causes a significant decline in the quality of life and a substantial economic burden on society. Yet, diagnosing cognitive impairment is apt to be missed or delayed due to its assessment being laborious. This study aimed to propose a new approach of classifying the risk of cognitive impairment in the elderly using sequential gait characteristics and machine learning techniques. A total of 108 community-dwelling elderly individuals participated in this study. The participants were categorized into three groups based on their scores of the mini-mental state examination. Each participant completed both the usual- and fast-paced walking on the straight path with two gyroscopes on each foot. By analyzing the foot sagittal angular velocity signals, the temporal gait parameters were extracted from each gait cycle. Five classical machine learning models and a long short-term memory network were respectively employed to produce the classifiers that used the time-consecutive temporal gait parameters as predictors of cognitive impairment. Five-fold cross-validation was applied to 70% of the data in each group, and the remaining 30% were used as test data. An F 1 -score of 0.974 was achieved in classifying the three groups by the long short-term memory network-based classifier that used the double-limb support, stance, step, and stride times at usual-paced walking and the double- and single-limb support, stance, and stride times at fast-paced walking as inputs. The proposed approach would pave the way for earlier diagnosis of cognitive impairment in non-clinical settings without professional help, which can facilitate more timely intervention.


  • Frailty Assessment Using Temporal Gait Characteristics and a Long Short-Term Memory Network

    by Dawoon Jung, Jinwook Kim, Miji Kim, Chang Won Won, Kyung-Ryoul Mun

    Faced with the rapidly aging world population, frailty has emerged as a major health burden among the elderly. This study aimed to investigate the feasibility of using temporal gait characteristics and a long short-term memory network for assessing frailty. Seventy-four community-dwelling elderly individuals participated in this study. The participants were categorized into three groups by their FRAIL scale: robust, pre-frail, and frail groups. The participants completed a 7-meter walking at the self-selected pace with a gyroscope on each foot. Analyzing the gyroscopic data produced seven temporal gait parameters per each gait cycle. Enumerating six consecutive values of each gait parameter produced the gait sequence features which were used as frailty predictors along with the demographic features. Five-fold cross-validation was applied to 70% of the data, and the remaining 30% were used as test data. An F 1 -score of 0.931 was achieved in classifying the robust, pre-frail, and frail groups by the random forest model trained with age, sex, and the outputs of the long short-term memory network-based classifier that used the initial and terminal double-limb support, step, and stride times as inputs. The proposed approach of assessing frailty using the arrhythmic gait pattern of the elderly and machine learning technique is novel and promising. Pioneering a way that self-monitor frailty at home without any help from experts, the study can contribute toearly diagnosis of frailty and make timely medical intervention possible.


  • Open Access

    An Empirical Study of Tokenization Strategies for Various Korean NLP Tasks

    by Kyubyong Park, Joohong Lee, Seongbo Jang, Dawoon Jung

    Typically, tokenization is the very first step in most text processing works. As a token serves as an atomic unit that embeds the contextual information of text, how to define a token plays a decisive role in the performance of a model.Even though Byte Pair Encoding (BPE) has been considered the de facto standard tokenization method due to its simplicity and universality, it still remains unclear whether BPE works best across all languages and tasks. In this paper, we test several tokenization strategies in order to answer our primary research question, that is, "What is the best tokenization strategy for Korean NLP tasks?" Experimental results demonstrate that a hybrid approach of morphological segmentation followed by BPE works best in Korean to/from English machine translation and natural language understanding tasks such as KorNLI, KorSTS, NSMC, and PAWS-X. As an exception, for KorQuAD, the Korean extension of SQuAD, BPE segmentation turns out to be the most effective.