Researching | Population-Depth Counting Algorithm Based on Multiscale Fusion

Laser & Optoelectronics Progress, Volume. 57, Issue 24, 241502(2020)

Population-Depth Counting Algorithm Based on Multiscale Fusion

Jing Zuo^* and Yulin Ba

Author Affiliations

School of Automation and Electrical Engineering, Lanzhou Jiaotong University, Lanzhou, Gansu 730070, China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (12)

References (24)

Paper Information

Figures & Tables(12)

Fig. 1. Crowding density map. (a) Original image; (b) geometric adaptive Gaussian kernel

Download full size

Fig. 2. Diagram of overall structure

Download full size

Fig. 3. Principle diagram of dilated convolution

Download full size

Fig. 4. Structure of MSB

Download full size

Fig. 5. Experimental results on ShanghaiTech dataset. (a) Original images; (b) ground-truth images; (c) estimated crowd density maps

Download full size

Fig. 6. Effect of feature fusion on training error and test loss. (a) Test loss; (b) training error

Download full size

Table 1. Parameters of model
View table
Table 1. Parameters of model
Parameter Content Parameter Content
Learning rate 0.001 Momentum 0.9
Optimizer Adma Normalization 0.001
Weight-decay 0.05 Batch-size 1

Table 2. Comparison of counting results on ShanghaiTech dataset

Table 2. Comparison of counting results on ShanghaiTech dataset

Algorithm	Part_A			Part_B
Algorithm	MAE	MSE		MAE		MSE
Algorithm in Ref.[18]	181.8	277.7	32.0		49.8
MCNN^[9]	110.2	173.2	26.4		41.3
SCNN^[20]	90.4	135	21.6		33.4
MSCNN^[21]	83.8	127.4	17.7		30.2
CSRNet^[10]	68.2	115.0	10.6		16.0
DADNet^[19]	64.2	99.9	8.8		13.5
Proposed algorithm	63.4	97.2	9.6		14.3

Table 3. Comparison of counting results on UCF_CC_50 dataset
View table
Table 3. Comparison of counting results on UCF_CC_50 dataset
Algorithm MAE MSE
Algorithm in Ref.[18] 467.0 498.5
MCNN^[9] 377.6 509.1
Algorithm in Ref.[22] 338.6 424.5
SCNN^[20] 318.1 439.2
DADNet^[19] 285.5 389.7
CSRNet^[10] 266.1 397.5
Proposed algorithm 257.2 380.8

Table 4. Accuracy comparing on WorldExpo'10 dataset%

Table 4. Accuracy comparing on WorldExpo'10 dataset%

Algorithm	Accuracy											Average accuracy
Algorithm	S1		S2		S3		S4		S5			Average accuracy
Algorithm in Ref.[18]		9.8		14.1		14.3		22.2		3.7			12.8
MCNN^[9]		3.4		20.6		12.9		13.0		8.1			11.6
SCNN^[20]		4.4		15.7		10.0		11.0		5.9			9.4
CP-CNN^[23]		2.9		14.7		10.5		10.4		5.8			8.86
CSRNet^[10]		2.9		11.5		8.6		16.6		3.4			8.6
RRSC^[24]		2.9		15.0		7.2		14.7		2.6			8.5
Proposed algorithm		2.6		15.3		9.8		9.4		4.7			8.36

Table 5. Comparison of model performance
View table
Table 5. Comparison of model performance
Algorithm Part _A Part_B
MAE MSE MAE MSE
No MSB 123.2 194.7 26.5 42.4
No feature fusion 70.5 119.4 11.3 18.7
Proposed algorithm 63.4 97.2 9.6 14.3

Table 6. Experimental results of transfer learning
View table
Table 6. Experimental results of transfer learning
Group MAE MSE
Group 1 290.3 458.7
Group 2 115.7 169.5

Tools

Get Citation

Copy Citation Text

Jing Zuo, Yulin Ba. Population-Depth Counting Algorithm Based on Multiscale Fusion[J]. Laser & Optoelectronics Progress, 2020, 57(24): 241502

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Apr. 20, 2020

Accepted: Jun. 1, 2020

Published Online: Nov. 18, 2020

The Author Email: Zuo Jing (1269132835@qq.com)

DOI:10.3788/LOP57.241502

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology