Multi‐dimensional weighted cross‐attention network in crowded scenes
Abstract Human detection in crowded scenes is one of the research components of crowd safety problem analysis, such as emergency warning and security monitoring platforms. Although the existing anchor‐free methods have fast inference speed, they are not suitable for object detection in crowded scene...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2021-12-01
|
| Series: | IET Image Processing |
| Subjects: | |
| Online Access: | https://doi.org/10.1049/ipr2.12298 |
| _version_ | 1798027610472579072 |
|---|---|
| author | Yefan Xie Jiangbin Zheng Xuan Hou Irfan Raza Naqvi Yue Xi Nailiang Kuang |
| author_facet | Yefan Xie Jiangbin Zheng Xuan Hou Irfan Raza Naqvi Yue Xi Nailiang Kuang |
| author_sort | Yefan Xie |
| collection | DOAJ |
| description | Abstract Human detection in crowded scenes is one of the research components of crowd safety problem analysis, such as emergency warning and security monitoring platforms. Although the existing anchor‐free methods have fast inference speed, they are not suitable for object detection in crowded scenes due to the model's inability to predict the well‐fined object detection bounding boxes. This work proposes an end‐to‐end anchor‐free network, Multi‐dimensional Weighted Cross‐Attention Network (MANet), which can perform real‐time human detection in crowded scenes. Specifically, the Double‐flow Weighted Feature Cascade Module (DW‐FCM) is used in the extractor to highlight the contribution of features at different levels. The Triplet Cross Attention Module (TCAM) is used in the detector head to enhance the association dependence of multi‐dimension features, further strengthening human boundary features' discrimination ability at a fine‐grained level. Moreover, the strategy of Adaptively Opposite Thrust Mapping (AOTM) ground‐truth annotation is proposed to achieve bias correction of erroneous mappings and reduce the iterations of useless learning of the network. These strategies effectively alleviate the defect that the existing anchor‐free network cannot correctly distinguish and locate the individual human in crowded scenes. Compared with the anchor‐based detection method, there is no need to set anchor parameters manually, and the detection speed can satisfy the real‐time application. Finally, through extensive comparative experiments on CrowdHuman and WIDER FACE datasets, the results demonstrate that the improved strategy achieves the state‐of‐the‐art result in the anchor‐free methods. |
| first_indexed | 2024-04-11T18:54:11Z |
| format | Article |
| id | doaj.art-7dad8689e7c14b8c9b305656b35d5263 |
| institution | Directory Open Access Journal |
| issn | 1751-9659 1751-9667 |
| language | English |
| last_indexed | 2024-04-11T18:54:11Z |
| publishDate | 2021-12-01 |
| publisher | Wiley |
| record_format | Article |
| series | IET Image Processing |
| spelling | doaj.art-7dad8689e7c14b8c9b305656b35d52632022-12-22T04:08:14ZengWileyIET Image Processing1751-96591751-96672021-12-0115143585359810.1049/ipr2.12298Multi‐dimensional weighted cross‐attention network in crowded scenesYefan Xie0Jiangbin Zheng1Xuan Hou2Irfan Raza Naqvi3Yue Xi4Nailiang Kuang5School of Computer Science and Engineering Northwestern Polytechnical University Xi'an PR ChinaSchool of Computer Science and Engineering Northwestern Polytechnical University Xi'an PR ChinaSchool of Computer Science and Engineering Northwestern Polytechnical University Xi'an PR ChinaSchool of Software Northwestern Polytechnical University Xi'an PR ChinaAeronautics Engineering College Air Force Engineering University of PLA Xi'an ChinaXi'an Microelectronics Technology Institute Xi'an PR ChinaAbstract Human detection in crowded scenes is one of the research components of crowd safety problem analysis, such as emergency warning and security monitoring platforms. Although the existing anchor‐free methods have fast inference speed, they are not suitable for object detection in crowded scenes due to the model's inability to predict the well‐fined object detection bounding boxes. This work proposes an end‐to‐end anchor‐free network, Multi‐dimensional Weighted Cross‐Attention Network (MANet), which can perform real‐time human detection in crowded scenes. Specifically, the Double‐flow Weighted Feature Cascade Module (DW‐FCM) is used in the extractor to highlight the contribution of features at different levels. The Triplet Cross Attention Module (TCAM) is used in the detector head to enhance the association dependence of multi‐dimension features, further strengthening human boundary features' discrimination ability at a fine‐grained level. Moreover, the strategy of Adaptively Opposite Thrust Mapping (AOTM) ground‐truth annotation is proposed to achieve bias correction of erroneous mappings and reduce the iterations of useless learning of the network. These strategies effectively alleviate the defect that the existing anchor‐free network cannot correctly distinguish and locate the individual human in crowded scenes. Compared with the anchor‐based detection method, there is no need to set anchor parameters manually, and the detection speed can satisfy the real‐time application. Finally, through extensive comparative experiments on CrowdHuman and WIDER FACE datasets, the results demonstrate that the improved strategy achieves the state‐of‐the‐art result in the anchor‐free methods.https://doi.org/10.1049/ipr2.12298Optical, image and video signal processingComputer vision and image processing techniquesMachine learning (artificial intelligence) |
| spellingShingle | Yefan Xie Jiangbin Zheng Xuan Hou Irfan Raza Naqvi Yue Xi Nailiang Kuang Multi‐dimensional weighted cross‐attention network in crowded scenes IET Image Processing Optical, image and video signal processing Computer vision and image processing techniques Machine learning (artificial intelligence) |
| title | Multi‐dimensional weighted cross‐attention network in crowded scenes |
| title_full | Multi‐dimensional weighted cross‐attention network in crowded scenes |
| title_fullStr | Multi‐dimensional weighted cross‐attention network in crowded scenes |
| title_full_unstemmed | Multi‐dimensional weighted cross‐attention network in crowded scenes |
| title_short | Multi‐dimensional weighted cross‐attention network in crowded scenes |
| title_sort | multi dimensional weighted cross attention network in crowded scenes |
| topic | Optical, image and video signal processing Computer vision and image processing techniques Machine learning (artificial intelligence) |
| url | https://doi.org/10.1049/ipr2.12298 |
| work_keys_str_mv | AT yefanxie multidimensionalweightedcrossattentionnetworkincrowdedscenes AT jiangbinzheng multidimensionalweightedcrossattentionnetworkincrowdedscenes AT xuanhou multidimensionalweightedcrossattentionnetworkincrowdedscenes AT irfanrazanaqvi multidimensionalweightedcrossattentionnetworkincrowdedscenes AT yuexi multidimensionalweightedcrossattentionnetworkincrowdedscenes AT nailiangkuang multidimensionalweightedcrossattentionnetworkincrowdedscenes |