The visual SLAM has been approved that it is extremely useful to obtain robust positions of camera for AR system. However, the traditional initialization of vSLAM will stop it being popular in AR system because the initialization approaches are too complex and inefficiently. For alleviating this limitation, we present a novel initializing method for monocular vSLAM system based on indoor model. In contrast to existing methods, the proposed method can even instantaneously initialize the vSLAM 3D map at the first frame. Given a single frame of the indoor image, the lines and vanishing points can be compiled, as well as the orientation map and a set of indoor model hypothesis, and the best fitting 3D indoor model can be estimated further. We use the indoor model to reliably initialize the vSLAM system. The experimental result on some public dataset proves the robustness and quickness of our initialization approach.
{"title":"Fast initialization method for monocular SLAM based on indoor model","authors":"Jisheng Huang, Ruyu Liu, Jianhua Zhang, Shengyong Chen","doi":"10.1109/ROBIO.2017.8324772","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324772","url":null,"abstract":"The visual SLAM has been approved that it is extremely useful to obtain robust positions of camera for AR system. However, the traditional initialization of vSLAM will stop it being popular in AR system because the initialization approaches are too complex and inefficiently. For alleviating this limitation, we present a novel initializing method for monocular vSLAM system based on indoor model. In contrast to existing methods, the proposed method can even instantaneously initialize the vSLAM 3D map at the first frame. Given a single frame of the indoor image, the lines and vanishing points can be compiled, as well as the orientation map and a set of indoor model hypothesis, and the best fitting 3D indoor model can be estimated further. We use the indoor model to reliably initialize the vSLAM system. The experimental result on some public dataset proves the robustness and quickness of our initialization approach.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133960346","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 : 2017-12-01DOI: 10.1109/ROBIO.2017.8324545
Zeyu Zeng, Yucheng He, Qi Zhang, Ying Hu, Yong Cao
In traditional endoscopic sinus surgery, the doctor has to hold the endoscopy in one hand, doing the surgery in the complicate nasal cavity with another hand holding the surgical instrument. Based on the nasal surgery scene people have developed many robot assisted sinus surgery (RASS) systems, which have provided surgeons stable surgical vision, improved the efficiency of surgery. However, most of these systems acquire operator to select multiple robot path points from the nostril to the surgical area (sinus), which is very time consuming and laborious. In this paper, we develop an improved A-star algorithm that can obtain path automatically only need to select the start point and target point. After registered four work coordinate spaces, we apply traditional A-star path planning algorithm with robot kinematic characteristics. Unlike previous approaches, trying to ensure the Remote Center of Motion (RCM) by design mechanical structure first, the control algorithm makes the endoscopy move in RCM mod, which meets the safety requirements. We have obtained promising results when we test our new approach.
{"title":"Approach and control for robot assisted sinus surgery","authors":"Zeyu Zeng, Yucheng He, Qi Zhang, Ying Hu, Yong Cao","doi":"10.1109/ROBIO.2017.8324545","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324545","url":null,"abstract":"In traditional endoscopic sinus surgery, the doctor has to hold the endoscopy in one hand, doing the surgery in the complicate nasal cavity with another hand holding the surgical instrument. Based on the nasal surgery scene people have developed many robot assisted sinus surgery (RASS) systems, which have provided surgeons stable surgical vision, improved the efficiency of surgery. However, most of these systems acquire operator to select multiple robot path points from the nostril to the surgical area (sinus), which is very time consuming and laborious. In this paper, we develop an improved A-star algorithm that can obtain path automatically only need to select the start point and target point. After registered four work coordinate spaces, we apply traditional A-star path planning algorithm with robot kinematic characteristics. Unlike previous approaches, trying to ensure the Remote Center of Motion (RCM) by design mechanical structure first, the control algorithm makes the endoscopy move in RCM mod, which meets the safety requirements. We have obtained promising results when we test our new approach.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134218417","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 : 2017-12-01DOI: 10.1109/ROBIO.2017.8324494
Zhihao Xu, Xuefeng Zhou, Taobo Cheng, Kezheng Sun, Dan Huang
In this paper, we consider the task-space tracking problem for robot manipulators with uncertain kinematics and dynamics. Imprecise kinematic parameters would cause errors in the solution of inverse kinematics, and the closed-loop system remains nonlinear and coupled. At the same time, task-space velocity or joint acceleration are usually required, which implies an increase of the production cost. Therefore, an adaptive control method is proposed, neither task-space velocity nor joint acceleration are needed. The measurement of task-space velocity is avoided using a low-pass filter, and by defining a second order reference trajectory, the joint acceleration is also eliminated. Using Lyapunov theory, we have proved that the end-effector tracking errors can asymptotically converge to zero. Examples and numeral simulations are provided to validate the effectiveness of the proposed tracking method.
{"title":"Adaptive task-space tracking for robot manipulators with uncertain kinematics and dynamics and without using acceleration","authors":"Zhihao Xu, Xuefeng Zhou, Taobo Cheng, Kezheng Sun, Dan Huang","doi":"10.1109/ROBIO.2017.8324494","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324494","url":null,"abstract":"In this paper, we consider the task-space tracking problem for robot manipulators with uncertain kinematics and dynamics. Imprecise kinematic parameters would cause errors in the solution of inverse kinematics, and the closed-loop system remains nonlinear and coupled. At the same time, task-space velocity or joint acceleration are usually required, which implies an increase of the production cost. Therefore, an adaptive control method is proposed, neither task-space velocity nor joint acceleration are needed. The measurement of task-space velocity is avoided using a low-pass filter, and by defining a second order reference trajectory, the joint acceleration is also eliminated. Using Lyapunov theory, we have proved that the end-effector tracking errors can asymptotically converge to zero. Examples and numeral simulations are provided to validate the effectiveness of the proposed tracking method.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133066565","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 : 2017-12-01DOI: 10.1109/ROBIO.2017.8324691
J. Kralik, D. Muldrew, D. Gunasekaran, Richard D. Lange
Human-level performance in robotic systems remains elusive. Therefore, an important complementary approach to developing intelligent systems is to use design principles of the brain. One major design element is the combination of multiple behavioral control systems. Although hierarchical architectures have been used in robotics, they have not closely mimicked those of the primate brain (including humans). We are developing a model of cognitive and action control in the human brain that captures key findings from the cognitive neuroscience literature. Here, we focus on the classic detour problem that requires reaching around barriers. We demonstrate model performance in both simulation and with a humanoid robot and show with the simulations that all levels in the model contribute to successful reaching, provided that the agent's environment poses problems of varying complexity.
{"title":"Cognitive and action control for goal-directed reaching in a humanoid robot","authors":"J. Kralik, D. Muldrew, D. Gunasekaran, Richard D. Lange","doi":"10.1109/ROBIO.2017.8324691","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324691","url":null,"abstract":"Human-level performance in robotic systems remains elusive. Therefore, an important complementary approach to developing intelligent systems is to use design principles of the brain. One major design element is the combination of multiple behavioral control systems. Although hierarchical architectures have been used in robotics, they have not closely mimicked those of the primate brain (including humans). We are developing a model of cognitive and action control in the human brain that captures key findings from the cognitive neuroscience literature. Here, we focus on the classic detour problem that requires reaching around barriers. We demonstrate model performance in both simulation and with a humanoid robot and show with the simulations that all levels in the model contribute to successful reaching, provided that the agent's environment poses problems of varying complexity.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133869896","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 : 2017-12-01DOI: 10.1109/ROBIO.2017.8324614
Guodong Yang, Dezheng Zhao, Wenkai Chang, E. Li, Zi-ze Liang
This paper presents a detection method for power transmission lines in low-resolution aerial image. The method is composed with two procedures as the line extraction procedure and the line fitting procedure. In the first procedure, an extraction algorithm based on Hessian matrix is proposed to segment the lines from the complex background. Besides, an image filter algorithm based on region growing is adopted to eliminate interference points caused by background. In the second procedure, a line fitting algorithm based on Expectation Maximization(EM) method is developed to get the geometry parameters and positions of the lines. Experiments show the effectiveness and efficiency of the proposed detection method.
{"title":"Detection for power transmission line in low-resolution aerial image","authors":"Guodong Yang, Dezheng Zhao, Wenkai Chang, E. Li, Zi-ze Liang","doi":"10.1109/ROBIO.2017.8324614","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324614","url":null,"abstract":"This paper presents a detection method for power transmission lines in low-resolution aerial image. The method is composed with two procedures as the line extraction procedure and the line fitting procedure. In the first procedure, an extraction algorithm based on Hessian matrix is proposed to segment the lines from the complex background. Besides, an image filter algorithm based on region growing is adopted to eliminate interference points caused by background. In the second procedure, a line fitting algorithm based on Expectation Maximization(EM) method is developed to get the geometry parameters and positions of the lines. Experiments show the effectiveness and efficiency of the proposed detection method.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122906447","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 : 2017-12-01DOI: 10.1109/ROBIO.2017.8324580
Zheng Huang, Diansheng Chen, Min Wang
With the development of society, the size and numbers of large-scale places such as shopping malls and airports continue to expand and the need for security automation increases. Patrol robot is a comprehensive system with environmental awareness, route planning, dynamic decision making, behavior decision making and so on [1]. The existing patrols are mainly inefficient manual patrols and it is a viable solution to make patrol robots achieve regular, fixed patrol or interrupted patrol in a complex environment. As a technology with excellent human-computer interaction, virtual reality can be applied to patrol robots and it can improve efficiency and reliability of patrol task. Based on the modeling of known patrol environment, 3 modules based on C++ programming language, which include virtual display module, registration module based on OpenCV visual library, robot remote control module based on TCP/IP protocol, were developed in the paper. This above 3 modules make the system achieve the remote control function and online patrol function of patrol robot. In addition, the realistic sense of the virtual reality technology and the accuracy of registration technology has been proved been proved by experiments.
{"title":"Design and application of intelligent patrol system based on virtual reality","authors":"Zheng Huang, Diansheng Chen, Min Wang","doi":"10.1109/ROBIO.2017.8324580","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324580","url":null,"abstract":"With the development of society, the size and numbers of large-scale places such as shopping malls and airports continue to expand and the need for security automation increases. Patrol robot is a comprehensive system with environmental awareness, route planning, dynamic decision making, behavior decision making and so on [1]. The existing patrols are mainly inefficient manual patrols and it is a viable solution to make patrol robots achieve regular, fixed patrol or interrupted patrol in a complex environment. As a technology with excellent human-computer interaction, virtual reality can be applied to patrol robots and it can improve efficiency and reliability of patrol task. Based on the modeling of known patrol environment, 3 modules based on C++ programming language, which include virtual display module, registration module based on OpenCV visual library, robot remote control module based on TCP/IP protocol, were developed in the paper. This above 3 modules make the system achieve the remote control function and online patrol function of patrol robot. In addition, the realistic sense of the virtual reality technology and the accuracy of registration technology has been proved been proved by experiments.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125691128","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 : 2017-12-01DOI: 10.1109/ROBIO.2017.8324769
M. Shahbazi, Navvab Kashiri, D. Caldwell, N. Tsagarakis
This paper introduces a computationally fast method for orientation trajectory planning in point-to-point motion tasks when the angular velocity and acceleration at the endpoints are constrained. Addressing such a problem with existing spherical-interpolation-based methods (in the context of unit quaternion) is not straightforward, since the inherent complexities of spherical curves necessitate task-specific tunings for satisfying all the boundary conditions. To tackle such a difficulty, we propound an interpolation function on the basis of standard polynomials in time with quaternion coefficients. We introduce a novel algorithm to determine varying polynomial coefficients in a way that the unit length of interpolated quaternion can be guaranteed. The performance of the developed planning algorithms is illustrated through a functional analysis and via simulation experiments on an anthropomorphic robotic arm. The results corroborate the merits of the presented approach especially in respecting arbitrary boundary conditions.
{"title":"Orientation planning in task space using quaternion polynomials","authors":"M. Shahbazi, Navvab Kashiri, D. Caldwell, N. Tsagarakis","doi":"10.1109/ROBIO.2017.8324769","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324769","url":null,"abstract":"This paper introduces a computationally fast method for orientation trajectory planning in point-to-point motion tasks when the angular velocity and acceleration at the endpoints are constrained. Addressing such a problem with existing spherical-interpolation-based methods (in the context of unit quaternion) is not straightforward, since the inherent complexities of spherical curves necessitate task-specific tunings for satisfying all the boundary conditions. To tackle such a difficulty, we propound an interpolation function on the basis of standard polynomials in time with quaternion coefficients. We introduce a novel algorithm to determine varying polynomial coefficients in a way that the unit length of interpolated quaternion can be guaranteed. The performance of the developed planning algorithms is illustrated through a functional analysis and via simulation experiments on an anthropomorphic robotic arm. The results corroborate the merits of the presented approach especially in respecting arbitrary boundary conditions.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125663245","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 : 2017-12-01DOI: 10.1109/ROBIO.2017.8324761
F. Wan, Zheng Wang, Brooke Franchuk, Xinyao Hu, Zhenglong Sun, Chaoyang Song
We describe a fluidic actuator design that replaces the sealed chamber of a hydraulic cylinder using a soft actuator to provide compliant linear compression with a large force (>100 N) at a low operation pressure (<50 kPa) for lower-limb wearable. The external shells constrain the deformation of the soft actuator under fluidic pressurization. This enables us to use latex party balloons as a quick and cheap alternative for initial design investigation. We found that the forces exerted by the soft material deformation are well-captured by the rigid shells, removing the necessity of explicitly describing the mechanics of the soft material deformation and its interaction with the rigid structure. One can use the classical Force, Pressure and Area formula factored with an efficiency parameter to characterize the actuator performance. Furthermore, we proposed an engineering design of the hybrid actuator using a customized soft actuator placed inside a single shell cavity with an open end for compression force. Our results show that the proposed design can generate a very high force within a short stroke distance. At a low input pressure of 50 kPa, the exerted block force is approaching only about 3% less than the classical equation predicted. The actuator is fitted to a new gait augmentation design for correcting knee alignment, which is usually challenging for actuators made from the purely soft material.
{"title":"Hybrid actuator design for a gait augmentation wearable","authors":"F. Wan, Zheng Wang, Brooke Franchuk, Xinyao Hu, Zhenglong Sun, Chaoyang Song","doi":"10.1109/ROBIO.2017.8324761","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324761","url":null,"abstract":"We describe a fluidic actuator design that replaces the sealed chamber of a hydraulic cylinder using a soft actuator to provide compliant linear compression with a large force (>100 N) at a low operation pressure (<50 kPa) for lower-limb wearable. The external shells constrain the deformation of the soft actuator under fluidic pressurization. This enables us to use latex party balloons as a quick and cheap alternative for initial design investigation. We found that the forces exerted by the soft material deformation are well-captured by the rigid shells, removing the necessity of explicitly describing the mechanics of the soft material deformation and its interaction with the rigid structure. One can use the classical Force, Pressure and Area formula factored with an efficiency parameter to characterize the actuator performance. Furthermore, we proposed an engineering design of the hybrid actuator using a customized soft actuator placed inside a single shell cavity with an open end for compression force. Our results show that the proposed design can generate a very high force within a short stroke distance. At a low input pressure of 50 kPa, the exerted block force is approaching only about 3% less than the classical equation predicted. The actuator is fitted to a new gait augmentation design for correcting knee alignment, which is usually challenging for actuators made from the purely soft material.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115844555","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 : 2017-12-01DOI: 10.1109/ROBIO.2017.8324416
Marc Bestmann, Florens Wasserfall, N. Hendrich, Jianwei Zhang
Most industrial robot workcells today are still characterized by a lot of external cables. In fact, cable management remains an important aspect of robot integration, often significantly impacting robot workspace and motion planning. In this paper, we discuss our solution to replace external cables with a wireless connection. While wireless networks are widely used for educational robots and teleoperation of mobile platforms, integration of wireless communication into industrial setups requires careful consideration, and the risks and potential loss of reliability must be evaluated. We present experimental results on the performance of our system, including measured Bluetooth latencies and packet-loss in several scenarios. In our case, the advantages of the cable-less system, including simplified motion planning, fewer self-collisions, and use of 3D-sensing for collision detection greatly outweigh any loss in reliability.
{"title":"Replacing cables on robotic arms by using serial via Bluetooth","authors":"Marc Bestmann, Florens Wasserfall, N. Hendrich, Jianwei Zhang","doi":"10.1109/ROBIO.2017.8324416","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324416","url":null,"abstract":"Most industrial robot workcells today are still characterized by a lot of external cables. In fact, cable management remains an important aspect of robot integration, often significantly impacting robot workspace and motion planning. In this paper, we discuss our solution to replace external cables with a wireless connection. While wireless networks are widely used for educational robots and teleoperation of mobile platforms, integration of wireless communication into industrial setups requires careful consideration, and the risks and potential loss of reliability must be evaluated. We present experimental results on the performance of our system, including measured Bluetooth latencies and packet-loss in several scenarios. In our case, the advantages of the cable-less system, including simplified motion planning, fewer self-collisions, and use of 3D-sensing for collision detection greatly outweigh any loss in reliability.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124512304","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 : 2017-12-01DOI: 10.1109/ROBIO.2017.8324596
Doudou Gao, Peijiang Yuan, Ning Sun, Xulei Wu, Ying Cai
In the background of big data and better computer hardware, convolutional neural network (CNN) has also made great progress. Since AlexNet model won the ILSVRC-2012 competition, CNN has been widely applied to computer version tasks including prediction task of face attributes. Different with single attribute prediction, multi-attribute task is subject to different attributes and is difficult to achieve particularly good performance. With the aim of building a CNN network with good performance and simple framework, ATNet is born as a based network with 4 convolution layers. Based on strong correlation, attributes have a positive effect on optimization of network weights, and conversely will be a negative effect. Therefore, 40 attributes of CelebA dataset are grouped to 3 groups. Because the correlation between groups is also different, groups are trained by branching from different depth. Meanwhile, 1×1 convolution layer is used to reduce dimension and limit the size of networks. Finally, we use a network named ATNet_GT with the above skills and get a good performance on CelebA dataset. The average accuracy of 40 attributes is 90.18%, and the standard deviation of accuracies is 7.25%.
{"title":"Face attribute prediction with convolutional neural networks","authors":"Doudou Gao, Peijiang Yuan, Ning Sun, Xulei Wu, Ying Cai","doi":"10.1109/ROBIO.2017.8324596","DOIUrl":"https://doi.org/10.1109/ROBIO.2017.8324596","url":null,"abstract":"In the background of big data and better computer hardware, convolutional neural network (CNN) has also made great progress. Since AlexNet model won the ILSVRC-2012 competition, CNN has been widely applied to computer version tasks including prediction task of face attributes. Different with single attribute prediction, multi-attribute task is subject to different attributes and is difficult to achieve particularly good performance. With the aim of building a CNN network with good performance and simple framework, ATNet is born as a based network with 4 convolution layers. Based on strong correlation, attributes have a positive effect on optimization of network weights, and conversely will be a negative effect. Therefore, 40 attributes of CelebA dataset are grouped to 3 groups. Because the correlation between groups is also different, groups are trained by branching from different depth. Meanwhile, 1×1 convolution layer is used to reduce dimension and limit the size of networks. Finally, we use a network named ATNet_GT with the above skills and get a good performance on CelebA dataset. The average accuracy of 40 attributes is 90.18%, and the standard deviation of accuracies is 7.25%.","PeriodicalId":197159,"journal":{"name":"2017 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"82 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134078818","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：