To solve the problem that the number of heads cannot be accurately extracted due to scale changes in crowd counting, a crowd density estimation based on multi-scale feature adaptive network was proposed(MSFANet). Firstly, deep convolution neural network is used to extract semantic features of different scales. Then, the expansion convolution network is introduced, and the expansion convolution in the scale-up unit is combined with the traditional convolution to further increase the receptive field and reduce the information loss caused by channel competition. Finally, channel attention is introduced to enhance the ability of multi-scale feature extraction and ensure the integrity of relevant and important information in the image. Experiments on datasets (ShanghaiTech, UCF_CC_50 and WorldExpo10) show that this algorithm is more accurate and robust than the current mainstream crowd counting algorithms.

Hui Gao, Miaolei Deng, Dexian Zhang, Wenjun Zhao. MSFANet: Crowd Density Estimation Based on Multi-scale Feature Adaptation. Academic Journal of Computing & Information Science (2021), Vol. 4, Issue 7: 52-58. https://doi.org/10.25236/AJCIS.2021.040708.

[1] Sabzmeydani P, Mori G. Detecting Pedestrians by Learning Shapelet Features [C]// IEEE Conference on Computer Vision & Pattern Recognition. IEEE, 2007.

[2] Chan A B, Vasconcelos N. Bayesian poisson regression for crowd counting [C]//2009 IEEE 12th International Conference on Computer Vision (ICCV),2009.

[3] Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks [J]. science, 2006, 313(5786): 504-507.

[4] Lecun Y, Bottou L. Gradient-based learning applied to document recognition [J]. Proceedings of the IEEE, 1998, 86(11):2278-2324.

[5] Zhang Y, Zhou D, Chen S, et al. Single-image crowd counting via multi-column convolutional neural network[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 589-597.

[6] Li Y, Zhang X, Chen D. CSRNet: Dilated convolutional neural networks for understanding the highly congested scenes [C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018.    

[7] Kang D, Chan A. Crowd Counting by Adaptively Fusing Predictions from an Image Pyramid [J]. arXiv preprint arXiv: 1805.06115, 2018. 

[8] X Jiang, Zhang L , Xu M , et al. Attention Scaling for Crowd Counting [C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020.

[9] Bai S, He Z, Qiao Y, et al. Adaptive Dilated Network With Self-Correction Supervision for Counting [C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020.

[10] Yang Y, Li G, Wu Z, et al. Reverse Perspective Network for Perspective-Aware Object Counting [C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020.

[11] Cao, Xinkun, et al. Scale Aggregation Network for Accurate and Efficient Crowd Counting [C]//european conference on computer vision, 2018: 757-773.

[12] Zhang C, Li H,Wang X,et al.Cross-scene crowd counting via deep convolutional neural networks [C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR),2015. 

[13] Zeiler M D,Fergus R.Visualizing and understanding convolutional networks [M] //Fleet D, Pajdla T, Schiele B, et al. Computer vision ECCV 2014. Lecture notes in computer science. Cham:Springer, 2014,8689:818-833.

[14] Sam D B, Surya S, Babu R V. Switching Convolutional Neural Network for Crowd Counting [C]// Computer Vision & Pattern Recognition. IEEE, 2017.

[15] Zan S, Yi X, Ni B, et al. Crowd Counting via Adversarial Cross-Scale Consistency Pursuit [C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2018.

[16] J. Gao, Q. Wang, and X. Li, “Pcc net: Perspective crowd counting via spatial convolutional network,” TCSVT, 2019.