• Laser & Optoelectronics Progress
  • Vol. 60, Issue 10, 1010001 (2023)
Haolin Liang, Huaiyu Cai*, Bochong Liu, Yi Wang, and Xiaodong Chen
Author Affiliations
  • Key Laboratory of Photoelectric Information, Ministry of Education, School of Precision Instruments and Optoelectronic Engineering, Tianjin University, Tianjin 300072, China
  • show less
    DOI: 10.3788/LOP213044 Cite this Article Set citation alerts
    Haolin Liang, Huaiyu Cai, Bochong Liu, Yi Wang, Xiaodong Chen. Road Falling Objects Detection Algorithm Based on Image and Point Cloud Fusion[J]. Laser & Optoelectronics Progress, 2023, 60(10): 1010001 Copy Citation Text show less
    Flow chart of algorithm for detecting falling objects
    Fig. 1. Flow chart of algorithm for detecting falling objects
    Structure diagram of ResNet-50
    Fig. 2. Structure diagram of ResNet-50
    ResNet-50 down-sampling module optimization. (a) Before optimization; (b) after optimization
    Fig. 3. ResNet-50 down-sampling module optimization. (a) Before optimization; (b) after optimization
    Schematic diagram of test vehicle and sensor installation
    Fig. 4. Schematic diagram of test vehicle and sensor installation
    Examples of experimental data collection
    Fig. 5. Examples of experimental data collection
    Extraction results of road objects. (a) Road edge extraction; (b) ground point cloud filtering; (c) ground point elimination; (d) point cloud clustering
    Fig. 6. Extraction results of road objects. (a) Road edge extraction; (b) ground point cloud filtering; (c) ground point elimination; (d) point cloud clustering
    Mapping result of point cloud region of interest in visual image
    Fig. 7. Mapping result of point cloud region of interest in visual image
    Detection result of scattered objects in a picture
    Fig. 8. Detection result of scattered objects in a picture
    Prediction result of scattered object image block
    Fig. 9. Prediction result of scattered object image block
    EquipmentModelParameter
    LidarVelodyne VLP-32C

    Maximum measuring range 200 m

    Field of view 360°(H)×40°(V,-25°-15°)

    Vertical resolution minimum 0.33°

    Horizontal resolution 0.1-0.4°

    CameraDaHua 5131M/CU210

    Resolution 1280×1024

    Pixel size 4.8 μm×4.8 μm

    Maximum frame rate 210 frame/s

    CMOS target surface size 1/2''

    GNSSMG910

    Centimeter-level positioning

    Maximum output frequency 20 Hz

    IMUIMU560Maximum output frequency 100 Hz
    Industrial PCDT-S2010MB-YH310MC4L

    CPU:Intel i7-8700

    RAM 16 GB

    2 TB mechanical hard disk+128 GB SSD

    Table 1. Main equipment parameters used by the system
    Total number of road objectsTotal number of falling objectsPredicted number of falling objectsNumber of falling objects predicted to be trueUnpredictable number of falling objectsPredicted number of falling objects that are falsePrecision /%Recall /%
    783260252239211394.8491.92
    Table 2. Experimental results of scattered objects detection algorithm
    Scattered object size /cmActual numberNumber of correctly detectedRecall /%
    <1017635.29
    10-3019118194.76
    >305252100.00
    Table 3. Detection of falling objects of different sizes
    Evaluation indexBefore optimizationAfter optimizationImprovement
    Precision /%93.4194.841.43
    Recall /%90.9091.921.02
    Table 4. Comparison of prediction results before and after network structure optimization
    Haolin Liang, Huaiyu Cai, Bochong Liu, Yi Wang, Xiaodong Chen. Road Falling Objects Detection Algorithm Based on Image and Point Cloud Fusion[J]. Laser & Optoelectronics Progress, 2023, 60(10): 1010001
    Download Citation