Fig. 1. Schematic of laser components
Fig. 2. Laser component structure diagram
Fig. 3. Haar features applied to pedestrian
Fig. 4. Rectangular feature
Fig. 5. Schematic of integral graph calculation
Fig. 6. Haar eigenvalue calculation diagram
Fig. 7. Training flowchart of Gentle Adaboost classifier
Fig. 8. Pedestrian detection test results
Fig. 9. Diagram of DSP pedestrian detection hardware circuit structure
Fig. 10. Diagram of STM32 circuit structure
Fig. 11. Automatic focusing control circuit board
Fig. 12. Flow of pedestrian detection program based on DSP
Fig. 13. Schematic of STM32 focusing program
Fig. 14. Test results of single person with different sizes
Fig. 15. Test results of multi persons
Target | Detection rate /% | False detection rate /% | Remarks |
---|
Pedestrians in the image | 88.6 | 9.3 | 200 images | Pedestrians in the video | 87.2 | 10.1 | 5 minutes video |
|
Table 1. Opencv pedestrian detection test results
Item | Speed beforeoptimization /ms | Speed afteroptimization /ms | Remark |
---|
Detection time | 4656 | 85 | 12 frame/s |
|
Table 2. Comparison of detection speed before and after optimization
Pedestrian height /pixel | Lens preset | Remarks |
---|
≥20 | 1 | For a single pedestrian, the pedestrian height is calculated as the size of a single pedestrian. When there are multiple pedestrians, the pedestrian height is calculated based on the largest of the multiple sizes. | ≥40 | 2 | ≥60 | 3 | ≥80 | 4 | ≥90 | 5 |
|
Table 3. Comparison table of target size and lens preset position