• Laser & Optoelectronics Progress
  • Vol. 58, Issue 20, 2010021 (2021)
Yuebo Meng1、2, Xuanrun Chen1, Guanghui Liu1、*, and Shengjun Xu1、2
Author Affiliations
  • 1College of Information and Control Engineering, Xi'an University of Architecture and Technology, Xi'an, Shaanxi 710055, China
  • 2Guangdong Artificial Intelligence and Digital Economy Laboratory (Guangzhou), Guangzhou, Guangdong 510000, China
  • show less
    DOI: 10.3788/LOP202158.2010021 Cite this Article Set citation alerts
    Yuebo Meng, Xuanrun Chen, Guanghui Liu, Shengjun Xu. Crowd Density Estimation Method Based on Multi-Feature Information Fusion[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2010021 Copy Citation Text show less
    Perspective structure of spatial attention
    Fig. 1. Perspective structure of spatial attention
    Multi-scale information aggregation structure
    Fig. 2. Multi-scale information aggregation structure
    Asymmetric convolution structure
    Fig. 3. Asymmetric convolution structure
    Diagram of feature fusion process
    Fig. 4. Diagram of feature fusion process
    Structural diagram of semantic embedding and feature fusion
    Fig. 5. Structural diagram of semantic embedding and feature fusion
    Structural diagram of multi-feature information fusion network
    Fig. 6. Structural diagram of multi-feature information fusion network
    Flow chart of multi-feature information fusion network algorithm
    Fig. 7. Flow chart of multi-feature information fusion network algorithm
    Experimental results in ShanghaiTech dataset. (a) Original graphs; (b) true-value graphs; (c) prediction results
    Fig. 8. Experimental results in ShanghaiTech dataset. (a) Original graphs; (b) true-value graphs; (c) prediction results
    Experimental results in Mall dataset. (a) Original graphs; (b) true-value graphs; (c) prediction results
    Fig. 9. Experimental results in Mall dataset. (a) Original graphs; (b) true-value graphs; (c) prediction results
    MethodPart_APart_B
    MAEMSEMAEMSE
    Algorithm in Ref.[33]181.8277.732.049.8
    MCNN[24]110.2173.226.441.3
    Switch-CNN[34]90.4135.021.633.4
    MSCNN[35]83.8127.417.730.2
    CSRNet[7]68.2115.010.616.0
    SANet[36]67.0104.58.413.6
    Proposed algorithm63.2102.88.012.8
    Table 1. Performance comparison among algorithms for ShanghaiTech dataset
    MethodMAEMSE
    Algorithm in Ref. [33]3.1515.7
    MCNN[24]2.217.33
    Switch-CNN[34]2.016.25
    MSCNN[35]2.127.04
    DecideNet[37]1.521.90
    Proposed algorithm1.431.72
    Table 2. Performance comparison among algorithms for Mall dataset
    MethodS1S2S3S4S5Average MAE
    MCNN[24]3.420.612.913.08.111.6
    MSCNN[35]7.815.414.911.85.811.7
    Switch-CNN[34]4.415.710.011.05.99.4
    DecideNet[37]2.013.148.917.44.759.23
    CSRNet[7]2.911.58.616.63.48.6
    Proposed algorithm2.611.28.914.23.68.1
    Table 3. Performance comparison among algorithms for Worldexpo’10 dataset
    MethodSize /MBAverage running speed of test image /s
    ShanghaiTechMallWorldexpo’10
    Algorithm in Ref. [33]7.12.360.321.22
    MCNN[24]19.22.310.321.15
    Switch-CNN[34]32.22.710.431.35
    MSCNN[35]22.22.340.321.14
    CSRNet[7]16.261.970.260.93
    Proposed algorithm (MSIA)17.392.110.291.02
    Proposed algorithm ( MSIA+PSA)21.42.320.311.11
    Table 4. Comparative analysis of algorithm complexity
    Yuebo Meng, Xuanrun Chen, Guanghui Liu, Shengjun Xu. Crowd Density Estimation Method Based on Multi-Feature Information Fusion[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2010021
    Download Citation