Table 7 Comparison of different methods for text-based traffic panel detection.
REF. | Dataset | Method | Precision (%) | Recall (%) | F1 score (%) | ACC (%) |
|---|---|---|---|---|---|---|
ASAYAR | CTPN | 0.67 | 0.85 | 0.75 | - | |
EAST | 0.71 | 0.89 | 0.79 | - | ||
TextBoxes++ | 0.44 | 0.63 | 0.52 | - | ||
ASAYAR | YOLOv5 | - | - | - | 0.99 | |
Tsinghua-Tencent 100 K (TT100 K) dataset | CNN + Deep Recurrent Model | 0.64 | 0.73 | 0.68 | - | |
ASAYAR | YOLOv5 | 0.99 | 0.96 | 0.95 | - | |
Persian text-based traffic panel | YOLOv3 | 0.97 | , 0.94 | 0.955 | - | |
traffic signs | MobileNetV2 | 0.93 | 0.91 | 0.92 | - | |
Jaguar Land Rover Research. | MSER and OCR | 0.94 | 0.98 | 0.95 | - | |
Kaggle dataset | TransNet | 0.97 | 0.97 | 0.97 | 0.97 | |
This study | ASAYAR | CNN + BiLSTM with Attention mechanism | 0.953 | 0.934 | 0.929 | 0.97 |
