Laser & Optoelectronics Progress, Volume. 57, Issue 6, 061010(2020)
Detection of Abnormal Escalator Behavior Based on Deep Neural Network
Figures & Tables(18)
Fig. 14. Detection results of three algorithms for large targets. (a) YOLOv3; (b) Tiny YOLOv3; (c) proposed algorithm
Fig. 15. Detection results of three algorithms for small targets. (a) YOLOv3; (b) Tiny YOLOv3; (c) proposed algorithm
Table 1. 18-layer deep separable convolution structure
View tableTable 1. 18-layer deep separable convolution structure
Type/stride Filter shape Output Conv dw/1 3×3×32 208×208×32 Conv/1 1×1×32×64 208×208×64 Conv dw/2 3×3×64 104×104×64 Conv/1 1×1×64×128 104×104×128 Conv dw/1 3×3×128 104×104×128 Conv/1 1×1×128×128 104×104×128 Conv dw/2 3×3×128 52×52×128 Conv/1 1×1×128×256 52×52×256 Conv dw/1 3×3×256 52×52×256 Conv/1 1×1×256×256 52×52×256 Conv dw/2 3×3×256 26×26×256 Conv/1 1×1×256×512 26×26×512 Conv dw/1 3×3×512 26×26×512 Conv/1 1×1×512×512 26×26×512 Conv dw/2 3×3×512 13×13×512 Conv/1 1×1×512×1024 13×13×1024 Conv dw/1 3×3×1024 13×13×1024 Conv/1 1×1×1024×1024 13×13×1024
Table 2. Width and height of priori box corresponding to different k values
View tableTable 2. Width and height of priori box corresponding to different k values
k=7 k=8 k=9 k=10 k=11 k=12 (62,61) (59,62) (60,59) (55,60) (56,60) (54,59) (81,117) (83,123) (75,105) (75,106) (75,106) (73,109) (123,172) (125,78) (100,147) (100,148) (100,148) (102,70) (140,83) (131,183) (136,81) (100,152) (122,72) (102,65) (186,255) (185,280) (146,205) (146,206) (140,209) (135,222) (231,146) (206,128) (192,293) (178,112) (177,120) (151,96) (287,323) (271,210) (210,133) (192,291) (189,312) (183,183) (278,372) (278,385) (242,162) (215,216) (188,323) (280,218) (266,398) (267,415) (230,131) (302,247) (264,414) (235,246) (315,258) (287,388) (323,210)
Table 3. Analysis of abnormal target detection performance for five different algorithms
View tableTable 3. Analysis of abnormal target detection performance for five different algorithms
Detection algorithm A /% FPS /(frame·s-1) F1 /% Faster RCNN 20.20 4.52 88.05 SSD 5.00 33.33 96.72 YOLOv3 5.50 25.64 95.63 Tiny YOLOv3 26.20 50.00 83.15 Proposed algorithm 3.40 43.48 97.60
Tools
Get Citation
Copy Citation Text
Xunsheng Ji, Bin Teng. Detection of Abnormal Escalator Behavior Based on Deep Neural Network[J]. Laser & Optoelectronics Progress, 2020, 57(6): 061010
Paper Information
Category: Image Processing
Received: Jul. 26, 2019
Accepted: Aug. 28, 2019
Published Online: Mar. 6, 2020
The Author Email: Bin Teng (2660087950@qq.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20