• Acta Optica Sinica
  • Vol. 38, Issue 12, 1215003 (2018)
Xia Hua1、*, Xinqing Wang1, Dong Wang1、2, Zhaoye Ma1, and Faming Shao1
Author Affiliations
  • 1 College of Field Engineering, PLA Army Engineering University, Nanjing, Jiangsu 210007, China
  • 2 Second Institute of Engineering Research and Design, Southern Theatre Command, Kunming, Yunnan 650222, China
  • show less
    DOI: 10.3788/AOS201838.1215003 Cite this Article Set citation alerts
    Xia Hua, Xinqing Wang, Dong Wang, Zhaoye Ma, Faming Shao. Multi-Objective Detection of Traffic Scenes Based on Improved SSD[J]. Acta Optica Sinica, 2018, 38(12): 1215003 Copy Citation Text show less
    Improved detection algorithm overall framework
    Fig. 1. Improved detection algorithm overall framework
    Example of CNN model extracting feature convolution kernels at various levels
    Fig. 2. Example of CNN model extracting feature convolution kernels at various levels
    Two-dimensional Gabor filter convolution kernel
    Fig. 3. Two-dimensional Gabor filter convolution kernel
    Three-dimensional Gabor filter convolution kernel
    Fig. 4. Three-dimensional Gabor filter convolution kernel
    Training process for optimal Gabor convolution kernel group
    Fig. 5. Training process for optimal Gabor convolution kernel group
    Mobile video target detection framework based on time-aware feature mapping
    Fig. 6. Mobile video target detection framework based on time-aware feature mapping
    Model processing video input and output schematics
    Fig. 7. Model processing video input and output schematics
    Example of M4 model detection results
    Fig. 8. Example of M4 model detection results
    ModelDatasetAP /%mAP /%Pf /%Pm /%Pd /%Pe /%
    PersonCarCyclist
    M0KITTI73.3671.5365.3270.0720.2119.3441.3219.13
    WD71.5969.6362.7567.9919.2521.3838.8320.54
    M1KITTI87.5382.1678.2882.6616.4817.9157.388.23
    WD85.6480.5974.3480.1918.9519.2851.4210.35
    M2KITTI77.1872.3568.6972.7412.3113.2957.8416.56
    WD73.5270.4564.8369.6115.1714.4952.4517.89
    M3KITTI88.4281.7374.3881.519.5311.6964.2514.53
    WD74.9272.3465.6370.9616.2415.1951.1617.41
    M4KITTI92.4292.2390.8591.835.197.1381.476.21
    WD88.4687.3883.2486.368.2611.2771.059.42
    Table 1. Comparison of model identification and detection effects
    MethodDatasetAP /%mAP /%Pd /%FPS /(frame·s-1)
    PersonCarCyclist
    Faster R-CNNKITTI83.2674.1375.4277.6145.2213.15
    WD81.4971.3368.6573.8236.6311.64
    DSOD300KITTI77.4372.2668.3872.6958.6858.23
    WD70.7369.3967.0469.0552.3250.35
    DSSD513KITTI75.4669.5368.3471.1159.4246.34
    WD72.1968.8366.4569.1649.7939.38
    YOLOv2 544KITTI79.4371.2567.3272.6660.8256.74
    WD73.2969.6368.8570.5954.8649.28
    M4KITTI92.4292.2390.8591.8381.4731.86
    WD88.4687.3883.2486.3671.0519.83
    Table 2. Comparison of detection and recognition with different algorithms
    Xia Hua, Xinqing Wang, Dong Wang, Zhaoye Ma, Faming Shao. Multi-Objective Detection of Traffic Scenes Based on Improved SSD[J]. Acta Optica Sinica, 2018, 38(12): 1215003
    Download Citation