• Laser & Optoelectronics Progress
  • Vol. 56, Issue 23, 231009 (2019)
Zhihua Qu**, Yiming Shao, Tianmin Deng*, Jie Zhu, and Xiaohua Song
Author Affiliations
  • School of Traffic and Transportation, Chongqing Jiaotong University, Chongqing 400074, China
  • show less
    DOI: 10.3788/LOP56.231009 Cite this Article Set citation alerts
    Zhihua Qu, Yiming Shao, Tianmin Deng, Jie Zhu, Xiaohua Song. Traffic Sign Detection and Recognition Under Complicated Lighting Conditions[J]. Laser & Optoelectronics Progress, 2019, 56(23): 231009 Copy Citation Text show less
    Schematic of MCT calculation process
    Fig. 1. Schematic of MCT calculation process
    Partial sample effects. (a) Original images; (b) transformed images
    Fig. 2. Partial sample effects. (a) Original images; (b) transformed images
    Overall flow chart of proposed algorithm
    Fig. 3. Overall flow chart of proposed algorithm
    Structure of multi-scale convolutional neural network
    Fig. 4. Structure of multi-scale convolutional neural network
    Partial sample pictures in GTSDB dataset
    Fig. 5. Partial sample pictures in GTSDB dataset
    Partial sample pictures in self-built dataset
    Fig. 6. Partial sample pictures in self-built dataset
    Detection results of proposed algorithm. (a) Detection results of GTSDB dataset; (b) detection results of CQD dataset; (c) detection results of CQN dataset
    Fig. 7. Detection results of proposed algorithm. (a) Detection results of GTSDB dataset; (b) detection results of CQD dataset; (c) detection results of CQN dataset
    DatasetAlgorithmTPFNFPPrecision /%Recall rate /%
    Proposed algorithm3545698.3398.61
    GTSDBMCT-Adaboost318384288.3389.33
    HOG+SVM3495798.0398.58
    Proposed algorithm583141098.3197.65
    CQDMCT-Adaboost519718585.9387.97
    HOG+SVM568381996.7693.73
    Proposed algorithm245181394.9693.16
    CQNMCT-Adaboost174354479.8283.25
    HOG+SVM206243386.1989.57
    Table 1. Detection performance of each algorithm
    DatasetAccuracy
    Proposed algorithmMulti-scale CNN[15]Random forests[16]LDA on HOG[17]
    GTSRB98.9498.3197.2095.68
    CQD98.3797.8895.2493.07
    CQN96.6191.7584.3767.33
    Table 2. Recognition accuracy of each algorithm%
    AlgorithmTime cost
    Proposed algorithm81
    MCT-Adaboost394
    HOG+SVM267
    Faster-RCNN102
    Table 3. Running time cost of each algorithmms
    Zhihua Qu, Yiming Shao, Tianmin Deng, Jie Zhu, Xiaohua Song. Traffic Sign Detection and Recognition Under Complicated Lighting Conditions[J]. Laser & Optoelectronics Progress, 2019, 56(23): 231009
    Download Citation