Acta Optica Sinica, Volume. 40, Issue 5, 0504001(2020)
Multi-Scale Infrared Pedestrian Detection Based on Deep Attention Mechanism
Fig. 1. Characteristic of pedestrian in U-FOV infrared images. (a) Large and medium scale pedestrians; (b) small scale pedestrians
Fig. 2. Architecture of multi-scale infrared pedestrian detection network based on Darknet53
Fig. 3. Architecture of attention module
Fig. 4. Residual module
Fig. 5. Principle of pedestrian detection
Fig. 6. Learning rate and loss curves. (a) Learning rate on Caltech dataset; (b) loss on Caltech dataset; (c) learning rate on U-FOV dataset; (d) loss on U-FOV dataset
Fig. 7. Salient coefficient and feature maps
Fig. 8. Distribution of pedestrian size in U-FOV test set
Fig. 9. Visualization results of infrared pedestrian detection
Fig. 10. P-R curves under different IoU thresholds. (a) IoU threshold is 0.3; (b) IoU threshold is 0.45; (c) IoU threshold is 0.5; (d) IoU threshold is 0.7
Fig. 11. Visualization results of infrared pedestrian detection on LTIR dataset at different scenes
|
|
Get Citation
Copy Citation Text
Bin Zhao, Chunping Wang, Qiang Fu, Yichao Chen. Multi-Scale Infrared Pedestrian Detection Based on Deep Attention Mechanism[J]. Acta Optica Sinica, 2020, 40(5): 0504001
Category: Detectors
Received: Sep. 23, 2019
Accepted: Nov. 27, 2019
Published Online: Mar. 10, 2020
The Author Email: Wang Chunping (wang_c_p@163.com)