Chinese Journal of Liquid Crystals and Displays, Volume. 38, Issue 12, 1707(2023)
Attention and cross-scale fusion for vehicle and pedestrian detection
Figures & Tables(14)
Fig. 7. Detection results comparion of SSD and proposed algorithm in the sub-samples of the PASCAL VOC
Fig. 8. Detection results comparion of SSD and proposed algorithm on RTP target data set
Table 1. Data distribution for RTP
View tableView in ArticleTable 1. Data distribution for RTP
数据来源 目标个数 Car Person 总计 104 785 27 216 KITTI 103 453 24 560 网络图像 708 1 650 真实街道拍摄 624 1 006
Table 2. Results of the ablation experiment
View tableView in ArticleTable 2. Results of the ablation experiment
SRBPN SAFM CCN mAP/% FPS 84.0 85.0 √ 85.5 80.6 √ √ 86.2 79.0 √ √ √ 87.1 77.0
Table 3. Comparison of detection accuracy of all types in the sub-samples of the PASCAL VOC dataset
View tableView in ArticleTable 3. Comparison of detection accuracy of all types in the sub-samples of the PASCAL VOC dataset
Model AP mAP Car Bus Bike Motor Person ION300 87.5 88.6 86.2 83.6 82.1 85.6 Faster R-CNN 84.7 83.1 79.0 77.5 76.7 80.2 YOLOv3［ 33 ］87.7 87.6 85.5 86.2 83.3 86.1 DC-SPP-YOLO［ 34 ］83.9 87.6 84.9 85.7 77.0 83.8 FCOS 88.1 84.8 86.8 84.8 84.2 85.7 LNFCOS 87.3 85.2 85.6 83.9 83.4 85.1 SSD300 85.7 87.0 83.9 84.0 79.4 84.0 DSSD321 86.2 85.6 84.9 86.7 79.7 84.6 RSSD［ 35 ］86.3 86.1 85.9 84.0 80.2 84.5 EDF-SSD 87.3 87.0 87.1 87.3 81.0 85.9 AFP-SSD［ 36 ］86.6 88.0 85.4 85.9 79.4 85.1 RLCADet［ 37 ］86.0 87.0 86.7 86.4 79.1 85.0 AFE-SSD［ 38 ］87.4 89.0 85.9 86.6 85.9 87.0 DF-SSD［ 39 ］86.3 85.4 85.0 86.2 79.1 84.4 Proposed 88.2 87.4 87.9 87.5 84.4 87.1
Table 4. Comparison of verification results on Titan X for different algorithms
View tableView in ArticleTable 4. Comparison of verification results on Titan X for different algorithms
Model Network mAP/% FPS Faster R-CNN VGG16 80.2 7.0 ION300 VGG16 85.6 - R-FCN［ 40 ］ResNet-101 84.1 9.0 DSSD321 ResNet-101 84.6 9.5 RSSD VGG16 84.5 35.0 SSD VGG16 84.0 46.0 Proposed VGG16 87.1 41.5
Table 5. Comparison of verification results on 1080Ti for different algorithms
View tableView in ArticleTable 5. Comparison of verification results on 1080Ti for different algorithms
Model Network mAP/% FPS YOLOv3 Darknet-53 86.1 39.0 RLCADet Resnet101 85.0 36.0 SSD VGG16 84.0 85.0 Proposed VGG16 87.1 77.0
Table 6. Comparison of detection results on RTP
View tableView in ArticleTable 6. Comparison of detection results on RTP
Model Network mAP/% FPS SSD VGG16 73.2 85.0 Proposed VGG16 77.1 77.0
Tools
Get Citation
Copy Citation Text
Jian-dong LI, Jia-qi LI, Hai-cheng QU. Attention and cross-scale fusion for vehicle and pedestrian detection[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(12): 1707
Paper Information
Category: Research Articles
Received: Feb. 6, 2023
Accepted: --
Published Online: Mar. 7, 2024
The Author Email: Jia-qi LI (dor_emma@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20