Pub Date : 2021-11-01DOI: 10.1109/INSAI54028.2021.00063
Yang Gong, P. Zhang
The demand for water quality in modern society is higher and higher, in order to quickly judge the water quality grade. This paper presents a water quality grade prediction model based on neural network. Firstly, the crawler technology is used to obtain the historical data of water quality monitoring; Then, the collected data are simply analyzed; Then, the neural network structure constructed by data training is used to continuously adjust the weight and bias parameters; Finally, the trained model is used to predict the water quality grade. After a lot of training and testing, the accuracy of the model in the training set can reach 97.30%; The accuracy rate in the test set can reach 96.66%, and good results have been achieved in both the training set and the test set. It has good generalization ability and can help predict the water quality level.
{"title":"Research and Implementation of Water Quality Grade Prediction based on Neural Network","authors":"Yang Gong, P. Zhang","doi":"10.1109/INSAI54028.2021.00063","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00063","url":null,"abstract":"The demand for water quality in modern society is higher and higher, in order to quickly judge the water quality grade. This paper presents a water quality grade prediction model based on neural network. Firstly, the crawler technology is used to obtain the historical data of water quality monitoring; Then, the collected data are simply analyzed; Then, the neural network structure constructed by data training is used to continuously adjust the weight and bias parameters; Finally, the trained model is used to predict the water quality grade. After a lot of training and testing, the accuracy of the model in the training set can reach 97.30%; The accuracy rate in the test set can reach 96.66%, and good results have been achieved in both the training set and the test set. It has good generalization ability and can help predict the water quality level.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"93 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":"125210972","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}
The registration of point cloud is essentially to obtain a relatively accurate coordinate transformation matrix through operation, and unify the point cloud data from multiview into the particular coordinate system through rigid transformations such as rotation and translation. Generally speaking, the registration is to discover the position conversion matrix of the overlap between clouds, which have an important effect in the domain of robot and computer vision. The purpose of this article is to comprehensively summarize the current progress of point cloud registration from two dimensions: algorithm optimization methods and deep learning methods. This paper first points out the possible application fields and development direction of point cloud registration in the future, then makes a comparison between different algorithms, and finally makes a proper analysis of the advantages and disadvantages of each algorithm.
{"title":"Registration of Point Clouds: A Survey","authors":"Dongfang Xie, Wei Zhu, Fengxiang Rong, Xu Xia, Huiliang Shang","doi":"10.1109/INSAI54028.2021.00034","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00034","url":null,"abstract":"The registration of point cloud is essentially to obtain a relatively accurate coordinate transformation matrix through operation, and unify the point cloud data from multiview into the particular coordinate system through rigid transformations such as rotation and translation. Generally speaking, the registration is to discover the position conversion matrix of the overlap between clouds, which have an important effect in the domain of robot and computer vision. The purpose of this article is to comprehensively summarize the current progress of point cloud registration from two dimensions: algorithm optimization methods and deep learning methods. This paper first points out the possible application fields and development direction of point cloud registration in the future, then makes a comparison between different algorithms, and finally makes a proper analysis of the advantages and disadvantages of each algorithm.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"38 3","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"120990021","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}
A power cable is a cable used to transmit and distribute high-power electrical energy in the backbone of the power system. Cross-linked polyethylene insulated cable has been widely used due to its various advantages, and its design life is generally 30 years, during which it will inevitably be affected by the internal or external environment of the cable, resulting in cable accidents. At present, most of the studies mainly focus on the local influencing factors of cables, which are broadly divided into three categories: mechanical properties, physicochemical properties, and electrical properties, and the results of the studies can only reflect the assessment of the influence factors on the degree of cable aging, and cannot specifically give the number of years the cable has been in operation. Therefore, this paper proposes an intelligent algorithm model based on BP neural network. The algorithm takes the measured operational data of the cable as input and then trains the network model to realize the calculation of the operational life of the cable. The simulation results show that the algorithm has the advantages of high accuracy and fast convergence.
{"title":"Cable Life Prediction Based on BP Neural Network","authors":"Yang Hu, Chizhi Huang, Dongdong Zhang, Chengxin Pang, Jinlong Wang, Chenhang Dong","doi":"10.1109/INSAI54028.2021.00055","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00055","url":null,"abstract":"A power cable is a cable used to transmit and distribute high-power electrical energy in the backbone of the power system. Cross-linked polyethylene insulated cable has been widely used due to its various advantages, and its design life is generally 30 years, during which it will inevitably be affected by the internal or external environment of the cable, resulting in cable accidents. At present, most of the studies mainly focus on the local influencing factors of cables, which are broadly divided into three categories: mechanical properties, physicochemical properties, and electrical properties, and the results of the studies can only reflect the assessment of the influence factors on the degree of cable aging, and cannot specifically give the number of years the cable has been in operation. Therefore, this paper proposes an intelligent algorithm model based on BP neural network. The algorithm takes the measured operational data of the cable as input and then trains the network model to realize the calculation of the operational life of the cable. The simulation results show that the algorithm has the advantages of high accuracy and fast convergence.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"28 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":"114437130","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/INSAI54028.2021.00067
Xun Liu, Yuying Li, NianQing Cai, W. Kuang, Guoqing Xia, Fangyu Lei
Existing popular methods for the recognition of plant leaf diseases with deep convolutional neural networks (DCNNs) improve the learning ability of traditional models by automatically learning the features of leaf images. However, these deep networks suffer from the concerns in terms of many parameters and high time complexity. To solve the limits, we propose a novel identification model (SCNN) of the plant leaf diseases based on shallow CNN. In SCNN, we reduce the number of parameters and the complexity by designing a new shallow network based on the deep learning technologies (BN and Dropout). Comprehensive evaluations on PlantVillage dataset demonstrate that our SCNN achieves state-of-the-art results.
{"title":"Recognition of Plant Leaf Diseases Based on Shallow Convolutional Neural Network","authors":"Xun Liu, Yuying Li, NianQing Cai, W. Kuang, Guoqing Xia, Fangyu Lei","doi":"10.1109/INSAI54028.2021.00067","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00067","url":null,"abstract":"Existing popular methods for the recognition of plant leaf diseases with deep convolutional neural networks (DCNNs) improve the learning ability of traditional models by automatically learning the features of leaf images. However, these deep networks suffer from the concerns in terms of many parameters and high time complexity. To solve the limits, we propose a novel identification model (SCNN) of the plant leaf diseases based on shallow CNN. In SCNN, we reduce the number of parameters and the complexity by designing a new shallow network based on the deep learning technologies (BN and Dropout). Comprehensive evaluations on PlantVillage dataset demonstrate that our SCNN achieves state-of-the-art results.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"42 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":"131820679","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/INSAI54028.2021.00057
Yijia Wu, Xinhua Zeng, Kaiqiang Feng, Donglai Wei, Lianghua Song
With the rapid development of Brain-Computer Interfaces (BCI), human visual decoding, as one of the important research directions of BCI, has aroused great attention. But most visual decoding researches focused on graphics decoding. In this paper, we investigate the possibility to build a new kind of BCI visual decoding based on visual color observation for the first time. We selected 10 subjects without color blindness disease to participate in our tests. They were asked to observe red, green, blue screens in turn with an interval of 1 second. 5 subjects took the test without a task, while another 5 subjects took the test with a task of simply counting one of the appearances of the color. The result shows that the visual color classification for group A without task can reach 83.57% on average, whereas the visual color classification for group B with the task is 78.57% on average. It shows that these subjects may distract themselves while taking the task, however, the classification accuracy is relatively higher than 66.11% for selected channels for both cases with or without taking a task as interference to BCI.
{"title":"Visual Color Decoding Using Brain-Computer Interfaces","authors":"Yijia Wu, Xinhua Zeng, Kaiqiang Feng, Donglai Wei, Lianghua Song","doi":"10.1109/INSAI54028.2021.00057","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00057","url":null,"abstract":"With the rapid development of Brain-Computer Interfaces (BCI), human visual decoding, as one of the important research directions of BCI, has aroused great attention. But most visual decoding researches focused on graphics decoding. In this paper, we investigate the possibility to build a new kind of BCI visual decoding based on visual color observation for the first time. We selected 10 subjects without color blindness disease to participate in our tests. They were asked to observe red, green, blue screens in turn with an interval of 1 second. 5 subjects took the test without a task, while another 5 subjects took the test with a task of simply counting one of the appearances of the color. The result shows that the visual color classification for group A without task can reach 83.57% on average, whereas the visual color classification for group B with the task is 78.57% on average. It shows that these subjects may distract themselves while taking the task, however, the classification accuracy is relatively higher than 66.11% for selected channels for both cases with or without taking a task as interference to BCI.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"23 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":"133011278","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/INSAI54028.2021.00041
Kun Yang, Jieyu Lin, Wei Ni, Lianghua Song
In recent years, deep learning algorithms have shown a trend towards larger models and larger datasets. Centralized training is unable keep up with the training requirements due to limited storage and computing resources, thus distributed learning is becoming an important area of research for improving learning efficiency. There are many studies on using the features of deep learning workload to design a central scheduler for production clusters.While existing work has been focusing on overall completion time and resource efficiency, little attention has been paid to the execution deadlines. To achieve a balance between the goals of deadline and non-deadline jobs, we design a Two-level Information-Agnostic Scheduling strategy(TIAS), which can schedule the two kinds of jobs together without knowing jobs’ training duration. In the first level, we use different priority calculation methods for the two kinds of jobs; in the second level, we design a new indicator "queue urgency" based on three observations to sort deadline jobs within the same queue. Experiments on a trace-driven simulator prove that TIAS can achieve the best trade-off between deadline miss rate and non-deadline jobs’ average job completion time(JCT) compared to existing solutions.
{"title":"TIAS: Two-level Information-Agnostic Job Scheduling in GPU Clusters","authors":"Kun Yang, Jieyu Lin, Wei Ni, Lianghua Song","doi":"10.1109/INSAI54028.2021.00041","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00041","url":null,"abstract":"In recent years, deep learning algorithms have shown a trend towards larger models and larger datasets. Centralized training is unable keep up with the training requirements due to limited storage and computing resources, thus distributed learning is becoming an important area of research for improving learning efficiency. There are many studies on using the features of deep learning workload to design a central scheduler for production clusters.While existing work has been focusing on overall completion time and resource efficiency, little attention has been paid to the execution deadlines. To achieve a balance between the goals of deadline and non-deadline jobs, we design a Two-level Information-Agnostic Scheduling strategy(TIAS), which can schedule the two kinds of jobs together without knowing jobs’ training duration. In the first level, we use different priority calculation methods for the two kinds of jobs; in the second level, we design a new indicator \"queue urgency\" based on three observations to sort deadline jobs within the same queue. Experiments on a trace-driven simulator prove that TIAS can achieve the best trade-off between deadline miss rate and non-deadline jobs’ average job completion time(JCT) compared to existing solutions.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"40 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":"133435313","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/INSAI54028.2021.00028
Shuai Huang, Dingkang Yang, Chuyi Zhong, Shi Yan, Lihua Zhang
In recent years, Ant Colony Optimization algorithm has become one of the most widely used heuristic algorithms and has been apply to solve different types of path planning problems. However, there still are some problems in Multi-Agent Path Finding, such as low convergence efficiency, easy to fall into local optimum and vertex conflict. In this paper, we proposed an Improved Ant Colony Optimization algorithm based on parameter optimization and vertex conflict resolution. First of all, we initialize the distribution of pheromones to reduce the blindness of the algorithm in the early stage. Secondly, we introduce an adaptive pheromone intensity and pheromone reduction factor to avoid the algorithm falling into local optimum. On this basis, the algorithmÿs global search ability and convergence speed are improved by dynamic modification of the evaporation factor and heuristic function. In addition, the strategy of dynamically modifying the influence factor and heuristic function improves the global search ability and convergence speed of the algorithm. To solve vertex conflict in MAPF, we use the design conflict prediction and resolution strategy to effectively avoid vertex conflict and improve the reliability of the multi-agent system. Simulation experiments verify the effectiveness and adaptability of IACO under different complexity environments, and prove that IACO has good convergence speed and path global optimization ability.
{"title":"An Improved Ant Colony Optimization Algorithm for Multi-Agent Path Planning","authors":"Shuai Huang, Dingkang Yang, Chuyi Zhong, Shi Yan, Lihua Zhang","doi":"10.1109/INSAI54028.2021.00028","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00028","url":null,"abstract":"In recent years, Ant Colony Optimization algorithm has become one of the most widely used heuristic algorithms and has been apply to solve different types of path planning problems. However, there still are some problems in Multi-Agent Path Finding, such as low convergence efficiency, easy to fall into local optimum and vertex conflict. In this paper, we proposed an Improved Ant Colony Optimization algorithm based on parameter optimization and vertex conflict resolution. First of all, we initialize the distribution of pheromones to reduce the blindness of the algorithm in the early stage. Secondly, we introduce an adaptive pheromone intensity and pheromone reduction factor to avoid the algorithm falling into local optimum. On this basis, the algorithmÿs global search ability and convergence speed are improved by dynamic modification of the evaporation factor and heuristic function. In addition, the strategy of dynamically modifying the influence factor and heuristic function improves the global search ability and convergence speed of the algorithm. To solve vertex conflict in MAPF, we use the design conflict prediction and resolution strategy to effectively avoid vertex conflict and improve the reliability of the multi-agent system. Simulation experiments verify the effectiveness and adaptability of IACO under different complexity environments, and prove that IACO has good convergence speed and path global optimization ability.","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":"114691638","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/INSAI54028.2021.00015
Xian Wang
At present, power distribution of 10kV and 6kV users generally adopts the wiring mode of connecting incoming and outgoing lines (distribution transformer) through the busbar. When each circuit breaker is equipped with protection, protection operation mode of three-stage series protection is formed, including incoming line protection, bus tie protection and outgoing line (distribution transformer) protection. With the development of the industrial process, users’ load is growing day by day. In particular, some smelting and foundry enterprises adopt two-stage power supply mode of switching station/distribution room. When each circuit breaker is equipped with protection, a six-stage series operation mode is formed. The protection operation mode is more complicated, as there are facts such as short protection circuit, similar fault currents, and no time limit for quick-break protection (all 0 seconds). Conventional protection devices have no selective protection functions and cannot achieve in-situ protection of circuit breakers in 10kV line. Local failures of the outgoing line can easily cause override trip of the incoming circuit breaker or the upper-level circuit breaker, which expands the power outage range, reduces reliability of power supply, and causes outage of some unnecessary equipment, thus affecting normal production of enterprises. In view of the above problems, this paper proposes a power distribution terminal device solution with selective trip protection function to realize intelligent protection, monitoring and management of 10kV substation and distribution network, and achieve multi-stage series protection of 10kV substation and distribution line against override trip and shutting. In the event of a line failure, it is possible to achieve in-situ quick protection, automatic quick isolation of the fault area, and automatic quick power supply transfer in the non-fault area, thereby greatly shortening power outage time, reducing power outage range, and increasing power supply reliability.
{"title":"Application of Equation of Light Shutting in Selective Protection Device","authors":"Xian Wang","doi":"10.1109/INSAI54028.2021.00015","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00015","url":null,"abstract":"At present, power distribution of 10kV and 6kV users generally adopts the wiring mode of connecting incoming and outgoing lines (distribution transformer) through the busbar. When each circuit breaker is equipped with protection, protection operation mode of three-stage series protection is formed, including incoming line protection, bus tie protection and outgoing line (distribution transformer) protection. With the development of the industrial process, users’ load is growing day by day. In particular, some smelting and foundry enterprises adopt two-stage power supply mode of switching station/distribution room. When each circuit breaker is equipped with protection, a six-stage series operation mode is formed. The protection operation mode is more complicated, as there are facts such as short protection circuit, similar fault currents, and no time limit for quick-break protection (all 0 seconds). Conventional protection devices have no selective protection functions and cannot achieve in-situ protection of circuit breakers in 10kV line. Local failures of the outgoing line can easily cause override trip of the incoming circuit breaker or the upper-level circuit breaker, which expands the power outage range, reduces reliability of power supply, and causes outage of some unnecessary equipment, thus affecting normal production of enterprises. In view of the above problems, this paper proposes a power distribution terminal device solution with selective trip protection function to realize intelligent protection, monitoring and management of 10kV substation and distribution network, and achieve multi-stage series protection of 10kV substation and distribution line against override trip and shutting. In the event of a line failure, it is possible to achieve in-situ quick protection, automatic quick isolation of the fault area, and automatic quick power supply transfer in the non-fault area, thereby greatly shortening power outage time, reducing power outage range, and increasing power supply reliability.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"12 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":"129168677","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/INSAI54028.2021.00027
Ming Wu, Chao Cheng, Huiliang Shang
Aiming at the problem that the current 2D Laser SLAM method can not take the high map accuracy and low computational complexity into account, this paper uses the 2D Graph-Based Laser SLAM algorithm. In the mapping stage, the idea of constructing a global map with submaps can effectively avoid the interference of moving objects in the environment; In the phase of pose optimization, the Gauss-Newton method [5] is used to find the new observation data of each frame, which is aligned to the optimal pose of the existing map, and then the observation data is updated to the map according to the pose; In the scan matching stage, the branch and bound algorithm is used to determine the robot's pose more quickly; In the navigation phase, DWA algorithm is used for local path planning. Through the experiments and comparison with Hector SLAM [3], we get better map and navigation results.
{"title":"2D LIDAR SLAM Based On Gauss-Newton","authors":"Ming Wu, Chao Cheng, Huiliang Shang","doi":"10.1109/INSAI54028.2021.00027","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00027","url":null,"abstract":"Aiming at the problem that the current 2D Laser SLAM method can not take the high map accuracy and low computational complexity into account, this paper uses the 2D Graph-Based Laser SLAM algorithm. In the mapping stage, the idea of constructing a global map with submaps can effectively avoid the interference of moving objects in the environment; In the phase of pose optimization, the Gauss-Newton method [5] is used to find the new observation data of each frame, which is aligned to the optimal pose of the existing map, and then the observation data is updated to the map according to the pose; In the scan matching stage, the branch and bound algorithm is used to determine the robot's pose more quickly; In the navigation phase, DWA algorithm is used for local path planning. Through the experiments and comparison with Hector SLAM [3], we get better map and navigation results.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"88 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":"116079973","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/INSAI54028.2021.00033
Fengxiang Rong, Dongfang Xie, Wei Zhu, Huiliang Shang, Liang Song
3D reconstruction is the process of obtaining the contour, color, depth, and other information of the real object through the sensor, and then transforming the 3D object in the real world into a 3D model that can be processed and displayed by the computer. According to the different sensors, there are a variety of ways to obtain data, among which the vision-based 3D reconstruction technology has been the focus of research in this field because of its wider scope of application and greater development prospects. At present, monocular, stereoscopic, and other reconstruction methods have been developed, and MVS (Multi View Stereo) is naturally developed in this process. This paper will start from sensor types, summarize the development process of 3d reconstruction methods, and focus on the introduction of multi-perspective stereo methods, including feature point matching in SFM (Structure from Motion) and SFM reconstruction method, traditional MVS method implementation and deep learning method implementation. Finally, the overall development of this field is summarized, and the future development is prospected.
三维重建是通过传感器获取真实物体的轮廓、颜色、深度等信息，然后将现实世界中的三维物体转化为可被计算机处理和显示的三维模型的过程。根据传感器的不同，获取数据的方式也多种多样，其中基于视觉的三维重建技术因其应用范围更广、发展前景更大，一直是该领域的研究热点。目前已经发展了单眼、立体等重建方法，MVS (Multi View Stereo)也在这个过程中自然发展起来。本文将从传感器类型入手，总结三维重建方法的发展历程，重点介绍多视角立体方法，包括SFM (Structure from Motion)中的特征点匹配和SFM重建方法、传统MVS方法的实现和深度学习方法的实现。最后对该领域的总体发展进行了总结，并对未来的发展进行了展望。
{"title":"A Survey of Multi View Stereo","authors":"Fengxiang Rong, Dongfang Xie, Wei Zhu, Huiliang Shang, Liang Song","doi":"10.1109/INSAI54028.2021.00033","DOIUrl":"https://doi.org/10.1109/INSAI54028.2021.00033","url":null,"abstract":"3D reconstruction is the process of obtaining the contour, color, depth, and other information of the real object through the sensor, and then transforming the 3D object in the real world into a 3D model that can be processed and displayed by the computer. According to the different sensors, there are a variety of ways to obtain data, among which the vision-based 3D reconstruction technology has been the focus of research in this field because of its wider scope of application and greater development prospects. At present, monocular, stereoscopic, and other reconstruction methods have been developed, and MVS (Multi View Stereo) is naturally developed in this process. This paper will start from sensor types, summarize the development process of 3d reconstruction methods, and focus on the introduction of multi-perspective stereo methods, including feature point matching in SFM (Structure from Motion) and SFM reconstruction method, traditional MVS method implementation and deep learning method implementation. Finally, the overall development of this field is summarized, and the future development is prospected.","PeriodicalId":232335,"journal":{"name":"2021 International Conference on Networking Systems of AI (INSAI)","volume":"59 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":"132905384","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：