Check out the research paper in the repository for a complete explanation of the algorithm!
| Detection | Images |
|---|---|
| Correct Detection | 754 |
| False Detection | 5 |
| No Detection | 2 |
| Partial Detection | 17 |
| Accuracy | 96.9% |
| Detection | Images |
|---|---|
| Correct | 774 |
| Acceptable | 177 |
| Incorrect | 58 |
| Accuracy | 94.1% |
(Note: "Acceptable" is defined as cases where the detection is off by one or two digits, but the timestamp's date can still be understood by a human. These cases arise in part due to error carrying forward from YOLOv3 localization.)
Averages around 3 seconds for localization and detection combined.
Thank you to the Systems Analysis and Forecasting Office team at the Ministry of Transportation for the support throughout the project. Another thank you to PyLessons for the amazing YOLOv3 program and the tutorials.