Article
Computer Science, Information Systems
Jingwei Dong, Ziqi Zhao, Tongxin Wang
Summary: The number of people in a crowd is crucial in various fields, and the accuracy of counting in public spaces is often compromised by uneven crowd distribution and differences in head scale due to varying distances from the camera. To address these issues, a deep learning crowd counting model called multi-scale dilated convolution networks (MSDCNet) is proposed, based on crowd density map estimation. The model consists of a front-end network, a core network, and a back-end network, all designed to extract features and improve counting accuracy. Experimental results on three public datasets demonstrate that the proposed model successfully solves these problems and outperforms representative models in terms of mean absolute error (MAE) and mean square error (MSE).
Article
Environmental Sciences
Bartosz Ptak, Dominik Pieczynski, Mateusz Piechocki, Marek Kraft
Summary: This study investigates crowd counting using low-altitude aerial images and evaluates neural network architectures to enhance image processing performance. Experiments show that input image resolution significantly impacts prediction quality, and careful consideration should be given before opting for a more complex neural network model.
Article
Computer Science, Information Systems
Donghua Liu, Guodong Wang, Guangtao Zhai
Summary: This paper proposes a crowd counting model based on a convolutional neural network to address the problem of head size variability. By fusing multi-scale high-level features, this method achieves better performance in the task of crowd counting. Experimental results on two public datasets demonstrate the effectiveness of the proposed approach.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Article
Chemistry, Multidisciplinary
Hui Gao, Miaolei Deng, Wenjun Zhao, Dexian Zhang
Summary: A new crowd counting method called SASNet is proposed, which focuses on estimating crowd density in population heterogeneous distribution. The method utilizes scene adaptive segmentation network and dual branches network to achieve stabilized performance and robustness.
APPLIED SCIENCES-BASEL
(2022)
Article
Computer Science, Artificial Intelligence
Zheng Xu, Deepak Kumar Jain, Pourya Shamsolmoali, Alireza Goli, Subramani Neelakandan, Amar Jain
Summary: Crowd counting and density estimation are crucial for ensuring public safety in surveillance videos. This study proposes a new approach called SMOHDL-CCA, which combines a Slime Mold Optimization algorithm with a Hybrid Deep Learning Enabled CC Approach. The proposed model accurately estimates the density map of crowded images and achieves comparable performance on standard datasets.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Lanjun Liang, Huailin Zhao, Fangbo Zhou, Qing Zhang, Zhili Song, Qingxuan Shi
Summary: This paper proposes an efficient scale-aware crowd counting network called SC2Net, which adopts an encoder-decoder framework and residual pyramid dilated convolution modules to extract multi-scale information and regress predicted density maps. Experimental results demonstrate the superiority of our proposed method over other state-of-the-art methods.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Lanjun Liang, Huailin Zhao, Fangbo Zhou, Mingyang Ma, Feng Yao, Xiaojun Ji
Summary: In this study, we propose a lightweight crowd counting network called PDDNet, which utilizes GhostNet to extract crowd features at the front-end and employs lightweight pyramidal convolution modules to extract multi-scale features at the back-end. The extracted features are then fed to transposed convolution layers to regress the crowd density map. Extensive experiments on commonly-used crowd counting datasets demonstrate the superiority of our model compared to state-of-the-art methods.
APPLIED INTELLIGENCE
(2023)
Article
Engineering, Civil
Jun Yi, Yiran Pang, Wei Zhou, Meng Zhao, Fujian Zheng
Summary: This paper proposes a novel perspective-embedded scale-selection multi-column network called PESSNet for crowd counting and high-quality density maps generation in congested urban transport systems. Experimental results demonstrate that PESSNet achieves reliable recognition performance and high robustness in different crowd counting tasks.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Computer Science, Information Systems
Liangjun Huang, Luning Zhu, Shihui Shen, Qing Zhang, Jianwei Zhang
Summary: This paper introduces a scale-aware representation learning network (SRNet) to address scale variation issues in dense crowd counting and crowd localization tasks. By encoding and decoding images and using two modules for multi-scale feature learning and spatial resolution enhancement, the SRNet has been proven effective in both qualitative and quantitative experiments.
Article
Computer Science, Software Engineering
Zhuangzhuang Miao, Yong Zhang, Yuan Peng, Haocheng Peng, Baocai Yin
Summary: Crowd counting is important for public security and urban management. Existing mainstream methods use convolutional neural networks (CNNs) to regress a density map but require detailed annotations, while weakly-supervised methods only need count annotations but often overlook the global perspective field and multi-level information. We propose DTCC, a weakly-supervised method that combines multi-level dilated convolution and transformer methods to achieve end-to-end crowd counting. Experimental results on four benchmark datasets show that DTCC outperforms other weakly-supervised methods and is comparable to fully-supervised methods.
COMPUTATIONAL VISUAL MEDIA
(2023)
Article
Biology
Shirsha Bose, Ritesh Sur Chowdhury, Rangan Das, Ujjwal Maulik
Summary: Biomedical image segmentation is crucial for medical image analysis, and deep learning algorithms allow for the design of advanced models to solve segmentation problems. The D3MSU-Net is proposed, which varies the receptive field at each level based on the resolution layer's depth and performs supervision at each resolution level. Experimental results demonstrate the superiority of the proposed network.
COMPUTERS IN BIOLOGY AND MEDICINE
(2022)
Article
Physiology
Zhensen Chen, Jieyun Bai, Yaosheng Lu
Summary: In this paper, a deep learning-based method for cardiac MRI segmentation is proposed, which addresses the issues of fuzzy boundaries and class ambiguity in existing methods by using dilated convolution network, edge fusion block, and directional feature maps.
FRONTIERS IN PHYSIOLOGY
(2023)
Article
Engineering, Electrical & Electronic
Ting Zhang, Huake Wang, Kaibing Zhang, Xingsong Hou
Summary: To address the scale variation of crowd, a deformable channel non-local network (DCNLNet) is proposed for crowd counting, which can simultaneously learn global context information and adaptive local receptive field. The proposed DCNLNet consists of two modules: deformable channel non-local block (DCNL) and spatial attention feature fusion block (SAFF). The DCNL encodes long-range dependencies and adaptive local correlation, benefiting for improving the spatial discrimination of features. The SAFF aims to aggregate cross-level information and learn specific weights for feature maps with spatial attention. Experimental results demonstrate that the proposed DCNLNet achieves compelling performance compared to other counting models.
ELECTRONICS LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Rui Wang, Reem Alotaibi, Bander Alzahrani, Arif Mahmood, Gaoxiang Wu, Han Xia, Abeer Alshehri, Sahar Aldhaheri
Summary: Intelligent crowd management is crucial for city monitoring, and the lack of labeled training samples poses a great challenge in crowd counting tasks. To enhance the performance of counting models, this paper proposes an automatic augmentation framework (AAC) based on deep reinforcement learning. Additionally, a large-scale crowd counting dataset HaCrowd is introduced. Experimental results demonstrate that AAC can generate flexible augmentation strategies for counting models tailored to specific datasets, leading to improved counting model performance.
Article
Computer Science, Artificial Intelligence
Xinyue Chen, Hua Yan, Tong Li, Jialang Xu, Fushun Zhu
Summary: Crowd counting faces challenges of scale variations and limited annotated data, while having similarities with object detection in attention areas. Existing methods often overlook these similarities and specialties. The proposed ASANet tackles these challenges with three branches and demonstrates outstanding performance on public datasets.