Implementation of Human Action Recognition System Using Multiple Kinect Sensors

Kwon, Beom; Kim, Doyoung; Kim, Junghwan; Lee, Sang Hoon; Kim, Jongyoo; Oh, Heeseok; Kim, Haksub

doi:10.1007/978-3-319-24075-6_32

Cited by 12 publications

(5 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the field of pattern recognition, the confusion matrix is widely used as one of the performance evaluation metrics [21]- [24]. In this study, to evaluate both classification and misclassification rates by comparing the predicted label with the ground truth, we computed the confusion matrix of each method as shown in Fig.…”

Section: E Resultsmentioning

confidence: 99%

Accurate Blind Lempel-Ziv-77 Parameter Estimation via 1-D to 2-D Data Conversion Over Convolutional Neural Network

2020

View full text Add to dashboard Cite

The data sequence compressed using the Lempel-Ziv-77 (LZ77) algorithm comprises a series of fixed-length tuples. To decompress this data, it is essential to know the encoding parameters such as the tuple length used for the compression. If this essential information becomes unavailable owing to conditions such as loss of the header, it is difficult to accurately determine the parameters using the compressed data. In this study, we investigate the blind estimation of tuple length from the LZ77-compressed data when the header is unavailable. To this end, we propose a novel idea of utilizing the LZ77 image generated from the LZ77-compressed data. The LZ77 image exhibits unique patterns based on the image size. The correlation between the image size and tuple length is indicated by different patterns of vertical lines in the LZ77 image. By exploiting a convolutional neural network (CNN), we develop an iterative algorithm while generating LZ77 images with different size. The results of the experiment on a public database show that the LZ77 image plays the role of an extremely powerful visual feature descriptor, and the proposed iterative algorithm estimates the tuple length with 100% accuracy.

show abstract

Section: E Resultsmentioning

confidence: 99%

Accurate Blind Lempel-Ziv-77 Parameter Estimation via 1-D to 2-D Data Conversion Over Convolutional Neural Network

2020

View full text Add to dashboard Cite

show abstract

“…Table 4 shows the F β score of each technique in all datasets. In this table, F β score is calculated using (16) and (17) as follows:…”

Section: B Performance Comparisonmentioning

confidence: 99%

Joint Swing Energy for Skeleton-Based Gender Classification

Kwon

Lee

2021

IEEE Access

Self Cite

View full text Add to dashboard Cite

Human gender classification in digital image content has received considerable attention from researchers for various applications, such as demographic research, video surveillance systems, and forensic science. In this study, we investigate three-dimensional (3D) human skeleton-based gender classification using a novel gait feature called joint swing energy (JSE). JSE is a kinematic gait feature that represents how distant a model skeleton's body joints are from anatomical planes while walking. However, anatomical planes are conventionally obtained from single static poses rather than dynamic motion. Therefore, in this study, we further investigate a novel method for obtaining transverse, frontal, and median planes from a 3D gait sequence. Using these planes, each joint's movement can be represented by coordinates centered on a human body rather than 3D Cartesian coordinates. Using the proposed methods, we extract JSEs of body joints from 3D gait sequences. We show that JSEs are different between walking men and women and propose the use of JSEs for machine classification of human gender. To demonstrate the effectiveness of the proposed JSE model on the gender classification task, we evaluate the performance of machine learning algorithms trained using JSE on four publicly available datasets, referred to as Datasets A, B, C, and D. Dataset A includes gait sequences from 164 persons between the ages of 17 and 45. Dataset B contains gait sequences from 104 persons between the ages of 17 and 36. In Dataset C, there are gait sequences for 30 persons between the ages of 23 and 55. In Dataset D, gait sequences for 30 persons between the ages of 21 and 55 are contained. The evaluation results demonstrate that the proposed technique achieves the highest classification accuracy and outperforms existing techniques for all datasets. These results suggest that human gender can be classified by JSEs extracted from the 3D gait sequence. INDEX TERMS Feature extraction, gait, gender classification, human skeleton, machine learning.

show abstract

“…The software development kit (SDK) of the Microsoft Kinect sensor was released in May 2011 [16]. Using the SDK, the user can acquire three-dimensional (3D) position data for 20 joints of the human body from the Kinect sensor [17][18][19]. In comparison with the human silhouette, the use of joint position enables view-invariant person identification.…”

Section: Introductionmentioning

confidence: 99%

Human Skeleton Data Augmentation for Person Identification over Deep Neural Network

Kwon

Lee

2020

Applied Sciences

Self Cite

View full text Add to dashboard Cite

With the advancement in pose estimation techniques, skeleton-based person identification has recently received considerable attention in many applications. In this study, a skeleton-based person identification method using a deep neural network (DNN) is investigated. In this method, anthropometric features extracted from the human skeleton sequence are used as the input to the DNN. However, training the DNN with insufficient training datasets makes the network unstable and may lead to overfitting during the training phase, causing significant performance degradation in the testing phase. To cope with a shortage in the dataset, we investigate novel data augmentation for skeleton-based person identification by utilizing the bilateral symmetry of the human body. To achieve this, augmented vectors are generated by sharing the anthropometric features extracted from one side of the human body with the other and vice versa. Thereby, the total number of anthropometric feature vectors is increased by 256 times, which enables the DNN to be trained while avoiding overfitting. The simulation results demonstrate that the average accuracy of person identification is remarkably improved up to 100% based on the augmentation on public datasets.

show abstract

Implementation of Human Action Recognition System Using Multiple Kinect Sensors

Cited by 12 publications

References 10 publications

Accurate Blind Lempel-Ziv-77 Parameter Estimation via 1-D to 2-D Data Conversion Over Convolutional Neural Network

Accurate Blind Lempel-Ziv-77 Parameter Estimation via 1-D to 2-D Data Conversion Over Convolutional Neural Network

Joint Swing Energy for Skeleton-Based Gender Classification

Human Skeleton Data Augmentation for Person Identification over Deep Neural Network

Contact Info

Product

Resources

About