Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00037
J. Qiu, F. Ye, Hua Zhou
As the scale of microprocessor chips and its design complexity continues to increase, the verification becomes more and more difficult. The microprocessor is the core component of computer system, and the instruction set of which is an important cornerstone for building the basic software and hardware ecosystem. The instruction set is a set of specifications for translating program language into machine language, and is the interface of software and hardware collaboration. This paper proposes an automatic, hierarchical verification platform and gives the verification results of the RISC-V base instruction. For the call of different instructions, only the top-level module name corresponding to the call needs to be changed.
{"title":"Automatic Verification Platform Based on RISC-V Architecture Microprocessor","authors":"J. Qiu, F. Ye, Hua Zhou","doi":"10.1109/INSAI54028.2021.00037","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00037","url":null,"abstract":"As the scale of microprocessor chips and its design complexity continues to increase, the verification becomes more and more difficult. The microprocessor is the core component of computer system, and the instruction set of which is an important cornerstone for building the basic software and hardware ecosystem. The instruction set is a set of specifications for translating program language into machine language, and is the interface of software and hardware collaboration. This paper proposes an automatic, hierarchical verification platform and gives the verification results of the RISC-V base instruction. For the call of different instructions, only the top-level module name corresponding to the call needs to be changed.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134105478","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00064
Liqin Shen, Mengmeng Zhang, Wentao Wang
As the key subsystem of rail transit vehicles, the reliability of train door subsystem directly affects the safety of vehicles' operation. Therefore, it is very necessary to obtain the operation state of train door in advance. The train door is a complex system that drives the movement of the train door leaf through the rotation of the motor driven screw rod. It is difficult to diagnose and predict its fault by establishing a mathematical model directly. It is also very difficult to collect a large number of sample data with fault labels for machine learning. Therefore, the expert system combined with domain expert knowledge is a better choice for train door fault diagnosis. By using mathematical expression as the rule of train door fault diagnosis expert system, this paper makes train door fault diagnosis more flexible, more scalable and easier to popularize.
{"title":"Application of Mathematical Expression Rules in Train Door Fault Diagnosis Expert System","authors":"Liqin Shen, Mengmeng Zhang, Wentao Wang","doi":"10.1109/INSAI54028.2021.00064","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00064","url":null,"abstract":"As the key subsystem of rail transit vehicles, the reliability of train door subsystem directly affects the safety of vehicles' operation. Therefore, it is very necessary to obtain the operation state of train door in advance. The train door is a complex system that drives the movement of the train door leaf through the rotation of the motor driven screw rod. It is difficult to diagnose and predict its fault by establishing a mathematical model directly. It is also very difficult to collect a large number of sample data with fault labels for machine learning. Therefore, the expert system combined with domain expert knowledge is a better choice for train door fault diagnosis. By using mathematical expression as the rule of train door fault diagnosis expert system, this paper makes train door fault diagnosis more flexible, more scalable and easier to popularize.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125996477","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Visual stimulus-based BCI system has received attention recently in the field of BCI. However, the existing visual EEG decoding methods are limited, so it is necessary to propose a new visual stimulus paradigm and dataset to study the new visual EEG decoding methods. In this paper, we contribute a real-world dataset containing new visual stimulus paradigm and propose two baseline algorithms for visual EEG decoding. Our dataset contains EEG data acquired from 9 subjects (age:22-27, 3 female) without dysopsia by using 64 channels wet electrode head-mounted BCI equipment. We get total of 2160 groups of data from all subjects. The raw data records EEG signals in response to two types of visual stimuli: One is a circle that varies from small to large, and the other varies from large to small. To prove the validity of the dataset, we use two kinds of machine learning algorithm for classification. By using SVM, the accuracy of a single subject is 65.32%~97.75% with an average of 76.72%. Through LSTM, the average accuracy achieves to 81.85%. In addition, we classify each channel separately and find the average accuracy of channels in the visual region (10 channels, 73.84%) is higher than that in the non-visual region (49 channels, 65.28%). Both methods demonstrate the validity of the dataset.
{"title":"A Visual EEG Paradigm and Dataset for Recognizing the Size Transformation of Images","authors":"Jingyi Liu, Kaiqiang Feng, Lianghua Song, Xinhua Zeng","doi":"10.1109/INSAI54028.2021.00040","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00040","url":null,"abstract":"Visual stimulus-based BCI system has received attention recently in the field of BCI. However, the existing visual EEG decoding methods are limited, so it is necessary to propose a new visual stimulus paradigm and dataset to study the new visual EEG decoding methods. In this paper, we contribute a real-world dataset containing new visual stimulus paradigm and propose two baseline algorithms for visual EEG decoding. Our dataset contains EEG data acquired from 9 subjects (age:22-27, 3 female) without dysopsia by using 64 channels wet electrode head-mounted BCI equipment. We get total of 2160 groups of data from all subjects. The raw data records EEG signals in response to two types of visual stimuli: One is a circle that varies from small to large, and the other varies from large to small. To prove the validity of the dataset, we use two kinds of machine learning algorithm for classification. By using SVM, the accuracy of a single subject is 65.32%~97.75% with an average of 76.72%. Through LSTM, the average accuracy achieves to 81.85%. In addition, we classify each channel separately and find the average accuracy of channels in the visual region (10 channels, 73.84%) is higher than that in the non-visual region (49 channels, 65.28%). Both methods demonstrate the validity of the dataset.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123966336","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00052
Haiyang Song, Xiaofeng Lu, Xuefeng Liu, Xiaoyu Zhu, Hewei Wang
The accuracy of the traditional head pose estimation method based on key feature points is easily affected by the accuracy of key feature points, serious occlusion or excessive angle deviation, resulting in bad deviation of the detection results. In order to improve the accuracy and stability of head pose estimation, a head pose estimation method using depth residual network ResNet101 as backbone network is proposed. The method AdaBound optimizer to optimize the training process gradient, use Softmax classifier and calculate the cross entropy loss function, and finally accurately predicts the head pose. We collected videos of stroke patients doing rehabilitation training, and established a new head posture data set after processing, which contains thousands of head posture RGB images of 40 stroke patients. We use the method proposed in this paper on this data set and the public dataset BIWI, and the results show that this method is very suitable for our dataset, and has good stability to different angles of the head posture, and has good robustness.
{"title":"Head Pose Estimation of Stroke Patients Based on Depth Residual Network","authors":"Haiyang Song, Xiaofeng Lu, Xuefeng Liu, Xiaoyu Zhu, Hewei Wang","doi":"10.1109/INSAI54028.2021.00052","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00052","url":null,"abstract":"The accuracy of the traditional head pose estimation method based on key feature points is easily affected by the accuracy of key feature points, serious occlusion or excessive angle deviation, resulting in bad deviation of the detection results. In order to improve the accuracy and stability of head pose estimation, a head pose estimation method using depth residual network ResNet101 as backbone network is proposed. The method AdaBound optimizer to optimize the training process gradient, use Softmax classifier and calculate the cross entropy loss function, and finally accurately predicts the head pose. We collected videos of stroke patients doing rehabilitation training, and established a new head posture data set after processing, which contains thousands of head posture RGB images of 40 stroke patients. We use the method proposed in this paper on this data set and the public dataset BIWI, and the results show that this method is very suitable for our dataset, and has good stability to different angles of the head posture, and has good robustness.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114370509","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00031
Na Yan, Xinhua Zeng, Lei Chen
With the development of artificial intelligence technology, emotion recognition has become an increasingly important research topic. Recognizing emotions only from the data with a single modality has its drawbacks. In this paper, the two modalities of facial expressions and EEG are integrated to realize the recognition of five types of emotions such as happiness, and the accuracy rate has reached a relatively satisfactory result. For facial expression modalities, this paper uses histogram equalization for preprocessing, then use LBP algorithm to extract facial expression features, and finally use SVM for expression recognition; for EEG modalities, this paper uses wavelet threshold denoising for preprocessing, and then use fractal dimension and multi-scale entropy algorithm to extract EEG signal features. This paper classifies EEG signals in the DEAP data set for emotion classification. Under the condition of using only one EEG channel FP1, the accuracy of SVM classification can reach 75.0%.
{"title":"Research on Emotion Recognition Based on Facial Expression and EEG","authors":"Na Yan, Xinhua Zeng, Lei Chen","doi":"10.1109/INSAI54028.2021.00031","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00031","url":null,"abstract":"With the development of artificial intelligence technology, emotion recognition has become an increasingly important research topic. Recognizing emotions only from the data with a single modality has its drawbacks. In this paper, the two modalities of facial expressions and EEG are integrated to realize the recognition of five types of emotions such as happiness, and the accuracy rate has reached a relatively satisfactory result. For facial expression modalities, this paper uses histogram equalization for preprocessing, then use LBP algorithm to extract facial expression features, and finally use SVM for expression recognition; for EEG modalities, this paper uses wavelet threshold denoising for preprocessing, and then use fractal dimension and multi-scale entropy algorithm to extract EEG signal features. This paper classifies EEG signals in the DEAP data set for emotion classification. Under the condition of using only one EEG channel FP1, the accuracy of SVM classification can reach 75.0%.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"PP 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126442905","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00050
Chao Cheng, M. Wu, Yuzhen Pan, Huiliang Shang
In the expansion of urban construction that emphasizes green and protecting the environment, emerging many green building technologies, but in this process, many waste building materials that are not conducive to recycling and may harm the health of workers, such as wooden boards with nails, are also produced. In this article, a set of robotic arm system based on the HSV algorithm is designed to handle these dangerous goods. particularly, we design an effective end-effector with rotating function for multiple scenarios. After theoretical analyzed, designed, and laboratory level verification, the system can achieve the expected functions well and has a good promotion and practical prospect in the construction of green cities.
{"title":"A Self-Designed Rotating End-Effector Based on Robotic System for Disposing of Nails in Wasted Board","authors":"Chao Cheng, M. Wu, Yuzhen Pan, Huiliang Shang","doi":"10.1109/INSAI54028.2021.00050","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00050","url":null,"abstract":"In the expansion of urban construction that emphasizes green and protecting the environment, emerging many green building technologies, but in this process, many waste building materials that are not conducive to recycling and may harm the health of workers, such as wooden boards with nails, are also produced. In this article, a set of robotic arm system based on the HSV algorithm is designed to handle these dangerous goods. particularly, we design an effective end-effector with rotating function for multiple scenarios. After theoretical analyzed, designed, and laboratory level verification, the system can achieve the expected functions well and has a good promotion and practical prospect in the construction of green cities.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133670643","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00049
Min Wang, Yucheng Fu, Rong Chu, Shouxian Zhu, Dahai Jing
In recent years, more and more attention has been paid to the automatic observation methods of ground-based cloud images. As it is related to real-time local weather forecasting, identifying the cloud type is always one of the basic observation items. However, due to the inability to extract subtle differences between classes, most of the existing automatic classification methods are not able to effectively recognize the cloud types defined by the World Meteorological Organization. Considering cloud images under ground-based scene have their own distinct characteristics, the proposed network architecture, called HACloudNet, exploits the informative features or classes selectively according to the vertical position of a pixel by introducing attention mechanism. We select ResNet18 as backbone network, adapt its structure to cloud classification, and combine it with the Height-driven Attention Layer, called HALayer, to guide the network to select more important features. Experiments on our ground-based scene dataset show that our method can significantly improve the performance of the backbone network. In particular, the accuracy of hard-to-classify samples has been obviously elevated. Comparison experiments show that our method is superior to the existing deep learning based cloud image classification methods without additional computational burden. It demonstrates that our method is more suitable for cloud image classification in real scenes.
{"title":"HACloudNet: A Ground-Based Cloud Image Classification Network Guided by Height-Driven Attention","authors":"Min Wang, Yucheng Fu, Rong Chu, Shouxian Zhu, Dahai Jing","doi":"10.1109/INSAI54028.2021.00049","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00049","url":null,"abstract":"In recent years, more and more attention has been paid to the automatic observation methods of ground-based cloud images. As it is related to real-time local weather forecasting, identifying the cloud type is always one of the basic observation items. However, due to the inability to extract subtle differences between classes, most of the existing automatic classification methods are not able to effectively recognize the cloud types defined by the World Meteorological Organization. Considering cloud images under ground-based scene have their own distinct characteristics, the proposed network architecture, called HACloudNet, exploits the informative features or classes selectively according to the vertical position of a pixel by introducing attention mechanism. We select ResNet18 as backbone network, adapt its structure to cloud classification, and combine it with the Height-driven Attention Layer, called HALayer, to guide the network to select more important features. Experiments on our ground-based scene dataset show that our method can significantly improve the performance of the backbone network. In particular, the accuracy of hard-to-classify samples has been obviously elevated. Comparison experiments show that our method is superior to the existing deep learning based cloud image classification methods without additional computational burden. It demonstrates that our method is more suitable for cloud image classification in real scenes.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125018883","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00018
Liguo Shi, Zhigen Xu, Yanzhen Li, Yang Hu
At present, the shore power connection in China is mainly completed by manual towing operation, which requires the mutual cooperation of wharf operators and mooring personnel. The operation has a large amount of labor, low efficiency, and poor working environment. With the development of science and technology and the improvement of industrial automation level, machine vision technology has been widely used in various fields. It is possible to use machine vision technology to replace manual connection of shore power cables. Therefore, in order to further improve the intelligent level of the port shore power system, solve the problem that the reverse power transmission operation depends on the manual dragging of the cable by the crew, improve the intelligent and automation level of the shore power collection system, and ensure the control and visual management of the ship shore power collection, the design of the shore power line and cable automatic assembly system based on machine vision is of great significance.
{"title":"Automatic Assembly System of Shore Connection Cable based on Machine Vision","authors":"Liguo Shi, Zhigen Xu, Yanzhen Li, Yang Hu","doi":"10.1109/INSAI54028.2021.00018","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00018","url":null,"abstract":"At present, the shore power connection in China is mainly completed by manual towing operation, which requires the mutual cooperation of wharf operators and mooring personnel. The operation has a large amount of labor, low efficiency, and poor working environment. With the development of science and technology and the improvement of industrial automation level, machine vision technology has been widely used in various fields. It is possible to use machine vision technology to replace manual connection of shore power cables. Therefore, in order to further improve the intelligent level of the port shore power system, solve the problem that the reverse power transmission operation depends on the manual dragging of the cable by the crew, improve the intelligent and automation level of the shore power collection system, and ensure the control and visual management of the ship shore power collection, the design of the shore power line and cable automatic assembly system based on machine vision is of great significance.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"127 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121752004","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00054
Jia Wu, Li Zhuo, Weijian Wu, Jinyue Qian, Jiantong Yue, Bailang Pan
This paper proposes a delaminating and districting demand response load fair and economical distribution method. The main methods are as follows: firstly, according to the power supply area and voltage level where the load is located, the vertical correspondence between users, distribution transformers, lines and substations is sorted out, and user attribute files are established to form a delaminating and districting load resource pool; then, based on load resource pool and power grid regulation demand, a comprehensive objective function considering user credit rating and regulation cost is established, and constraints such as regulation demand, recovery time, duration and regulation range are introduced to form a demand response load distribution model. Finally, the mixed integer programming technology is used to solve the model, so as to obtain the load demand response scheme that meets the principle of fair and economic distribution. Example test results verify the correctness and effectiveness of this method.
{"title":"Load Allocation Method Based on Fairness and Economy in Hierarchical and District Demand Response","authors":"Jia Wu, Li Zhuo, Weijian Wu, Jinyue Qian, Jiantong Yue, Bailang Pan","doi":"10.1109/INSAI54028.2021.00054","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00054","url":null,"abstract":"This paper proposes a delaminating and districting demand response load fair and economical distribution method. The main methods are as follows: firstly, according to the power supply area and voltage level where the load is located, the vertical correspondence between users, distribution transformers, lines and substations is sorted out, and user attribute files are established to form a delaminating and districting load resource pool; then, based on load resource pool and power grid regulation demand, a comprehensive objective function considering user credit rating and regulation cost is established, and constraints such as regulation demand, recovery time, duration and regulation range are introduced to form a demand response load distribution model. Finally, the mixed integer programming technology is used to solve the model, so as to obtain the load demand response scheme that meets the principle of fair and economic distribution. Example test results verify the correctness and effectiveness of this method.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126445867","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00039
Ziming Wang, Yang Lu, Wei Ni, Liang Song
With depth information more easily accessible even on mobile devices, leveraging RGB and depth information for RGB-D training provides a new way to enhance human pose estimation performance. In this paper, we propose an RGB-D based approach for human pose estimation. The main contributions of this paper are: 1) improving the accuracy and robustness of the model by utilizing depth image, 2) establishing a lightweight network architecture to improve the performance in detection speed, which makes it suitable for deployment on mobile devices. Qualitative and quantitative analyses on experimental results demonstrate that our model outperforms Open-Pose by 34% in detection speed, reduces model size to 42% at the same time. Our model also provides some advantages in specific background environments.
{"title":"An RGB-D Based Approach for Human Pose Estimation","authors":"Ziming Wang, Yang Lu, Wei Ni, Liang Song","doi":"10.1109/INSAI54028.2021.00039","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00039","url":null,"abstract":"With depth information more easily accessible even on mobile devices, leveraging RGB and depth information for RGB-D training provides a new way to enhance human pose estimation performance. In this paper, we propose an RGB-D based approach for human pose estimation. The main contributions of this paper are: 1) improving the accuracy and robustness of the model by utilizing depth image, 2) establishing a lightweight network architecture to improve the performance in detection speed, which makes it suitable for deployment on mobile devices. Qualitative and quantitative analyses on experimental results demonstrate that our model outperforms Open-Pose by 34% in detection speed, reduces model size to 42% at the same time. Our model also provides some advantages in specific background environments.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"150 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116629859","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：