Fig 1.
Block diagram of proposed method.
Fig 2.
Example of the Byte-MCT code generation.
Table 1.
AdaBoost learning algorithm based on landmarks.
Fig 3.
Example of landmarks in a traffic sign using the proposed AdaBoost learning algorithm.
Fig 4.
Examples of sliding-window search and parallel-window search.
Table 2.
Example of GPU kernel function.
Fig 5.
Example of verification of feature extraction.
Fig 6.
Example of entire sign recognition structure.
Fig 7.
Example of proposed model architecture.
Fig 8.
Examples from the LISA US speed-limit dataset and our real-world driving dataset.
Table 3.
Specification of test-sets.
Table 4.
Results of detection performance using ACF without classification.
Table 5.
Results of detection and recognition performance using the proposed method.
Table 6.
Results of detection and recognition performance using the single CNN.
Fig 9.
Resultant examples of proposed method for the LISA speed-limit test-set.
Fig 10.
Resultant examples of proposed method for the Germany (DE-D) test-set.
Fig 11.
Resultant examples of proposed method for the Korea (KR-D) test-set.
Fig 12.
Incorrect results using the proposed method for the Korea (KR-D) test-set.
Table 7.
Results of detection performance comparison.
Table 8.
Comparison of time consumption using CPU or GPGPU.