Table 5.
Performance comparison of the proposed YOLOv3-Human model with other methods on the same night databases.
| Method | Database | Purpose | Detection Time | Accuracy (AP Score) |
|---|---|---|---|---|
| TIRFaceNet model | DHU dataset | Facial recognition | 89.7% | |
| YOLO model | FLIR dataset | Person detector for small and multiple subjects | - | 29.36% |
| MMTOD model | KAIST dataset | Person detector for small and multiple subjects | - | 49.39 |
| FLIR dataset | - | 54.69 | ||
| YOLOv3-Human model | KAIST dataset | Person detector for small and multiple subjects | 10 ms | 65.01% |
| FLIR dataset | 8 ms | 67.54% | ||
| DHU Night dataset | Integrated face and gait recognition | 7 ms | 99% for face and gender recognition 90% for gait recognition |