• Laser & Optoelectronics Progress
  • Vol. 59, Issue 12, 1215017 (2022)
Yuemeng Zhao1、2 and Huigang Liu1、2、*
Author Affiliations
  • 1Engineering Research Center of Thin Film Optoelectronics Technology, Ministry of Education, College of Electronic Information and Optical Engineering, Nankai University, Tianjin 300350, China
  • 2Tianjin Key Laboratory of Optoelectronic Sensor and Sensing Network Technology, Tianjin 300350, China
  • show less
    DOI: 10.3788/LOP202259.1215017 Cite this Article Set citation alerts
    Yuemeng Zhao, Huigang Liu. Detection and Tracking of Low-Altitude Unmanned Aerial Vehicles Based on Optimized YOLOv4 Algorithm[J]. Laser & Optoelectronics Progress, 2022, 59(12): 1215017 Copy Citation Text show less
    Schematic diagram of YOLOv4 structure
    Fig. 1. Schematic diagram of YOLOv4 structure
    Comparsion between Darknet-53 and CSPDarknet-53. (a) Darknet-53; (b) CSPDarknet-53
    Fig. 2. Comparsion between Darknet-53 and CSPDarknet-53. (a) Darknet-53; (b) CSPDarknet-53
    Cascade matching process
    Fig. 3. Cascade matching process
    K-means clustering analysis result
    Fig. 4. K-means clustering analysis result
    Optimized YOLOv4 network structure
    Fig. 5. Optimized YOLOv4 network structure
    Detection and tracking model based on a combination of optimized YOLOv4 and DeepSORT
    Fig. 6. Detection and tracking model based on a combination of optimized YOLOv4 and DeepSORT
    Loss function curve
    Fig. 7. Loss function curve
    Test results. (a) YOLOv4; (b) optimized YOLOv4
    Fig. 8. Test results. (a) YOLOv4; (b) optimized YOLOv4
    Visual detection effects of small targets. (a) YOLOv3; (b) YOLOv4; (c) optimized YOLOv4
    Fig. 9. Visual detection effects of small targets. (a) YOLOv3; (b) YOLOv4; (c) optimized YOLOv4
    Visual tracking performance of optimized YOLOv4-DeepSORT method
    Fig. 10. Visual tracking performance of optimized YOLOv4-DeepSORT method
    Feature LevelSize of feature mapValue of anchor
    Feature Level 119×19(485,493),(495,279)
    Feature Level 238×38(128,110),(241,186)
    Feature Level 376×76(36,75),(76,55)
    Feature Level 4152×152(12,16),(19,36)
    Table 1. Value of anchor in LARotorcraft dataset
    Detection algorithmAP /%AP50 /%AP75 /%Detection speed /(frame·s-1
    YOLOv36977.141.58.9
    YOLOv475.693.286.512.7
    Optimized YOLOv477.297.190.110.8
    Table 2. Performance comparison of different detection algorithms
    Yuemeng Zhao, Huigang Liu. Detection and Tracking of Low-Altitude Unmanned Aerial Vehicles Based on Optimized YOLOv4 Algorithm[J]. Laser & Optoelectronics Progress, 2022, 59(12): 1215017
    Download Citation