Three-Dimensional Human Hand Pose Estimation Based on Finger-Point Reinforcement and Multi-Level Feature Fusion

Kaiyi Zhang; Ru Hong; Shaoyan Gai; Feipeng Da

doi:10.3788/AOS202242.1915001

Acta Optica Sinica, Volume. 42, Issue 19, 1915001(2022)

Three-Dimensional Human Hand Pose Estimation Based on Finger-Point Reinforcement and Multi-Level Feature Fusion

Kaiyi Zhang^1,2, Ru Hong^1,2, Shaoyan Gai^1,2, and Feipeng Da^1,2,3、*

Author Affiliations

¹School of Automation, Southeast University, Nanjing 210096, Jiangsu , China

²Key Laboratory of Measurement and Control of Complex Engineering Systems, Ministry of Education, Southeast University, Nanjing 210096, Jiangsu , China

³Shenzhen Research Institute, Southeast University, Shenzhen 518036, Guangdong , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Overview of overall algorithm flow

Download full size

Fig. 2. Comparison of three groups of gestures (black dots on top are parts with higher confidence parameters, and original point clouds are shown on bottom)

Download full size

Fig. 3. SE module and MFSE module. (a) SE module; (b) MFSE module

Download full size

Fig. 4. Error of each finger joint point by different methods

Download full size

Fig. 5. Proportion of test frames with errors within different thresholds

Download full size

Fig. 6. Comparison of experimental results under ICVL dataset

Download full size

Fig. 7. Comparison of experimental results under MSRA dataset

Download full size

Table 1. Error distance of each joint of FPR strategy on MSRA dataset

View table

Table 1. Error distance of each joint of FPR strategy on MSRA dataset

Joint	Baseline	Baseline+FPR
Mean error	8.5530	8.1930
Palm Index_R Index_T Mid_R Mid_T Ring_R Ring_T	8.6665 6.8979 11.1541 5.4392 10.9231 5.7619 9.7697	8.2581 6.4721 10.5203 5.1762 10.3742 5.5396 9.4127
Pinky_R	7.4491	7.1871
Pinky_T Thumb_R Thumb_T	10.2487 7.5354 13.9441	9.6389 7.2211 13.7389

Table 2. Error distance of each joint on MSRA dataset by different methods

View table

Table 2. Error distance of each joint on MSRA dataset by different methods

Joint	Baseline	Baseline+SE	Baseline+MFSE
Mean error	8.5320	8.3280	8.2050
Palm	8.6665	8.4958	8.3172
Index_R	6.8979	6.5829	6.5038
Index_T	11.1541	10.6960	10.4891
Mid_R	5.4392	5.2140	5.1937
Mid_T	10.9231	10.2835	10.1443
Ring_R	5.7619	5.6199	5.5399
Ring_T	9.7697	9.4182	9.2983
Pinky_R	7.4491	7.3386	7.1515
Pinky_T	10.2487	10.1666	9.7651
Thumb_R	7.5354	7.4101	7.2387
Thumb_T	13.9441	13.9072	13.7871

Table 3. Average error distance of each method on MSRA and ICVL datasets

View table

Table 3. Average error distance of each method on MSRA and ICVL datasets

Method	Mean error in MSRA	Mean error in ICVL
Hand PointNet^［12］	8.505	6.935
3D DenseNet^［21］	7.98	6.77
SHPR-NET^［22］	7.96	7.2
CNN Model^［23］	8.3	7.1
Bayesian DeepPrior^［25］		10.1
Pose REN^［24］	8.6	6.79
SO-HandNet^［13］		7.7
PCHPS^［26］	7.117	8.893
PointNet+MFSE	8.203	6.854
PointNet+FPR	8.192	6.728
PointNet+MFSE+FPR	7.942	6.673

Tools

Get Citation

Copy Citation Text

Kaiyi Zhang, Ru Hong, Shaoyan Gai, Feipeng Da. Three-Dimensional Human Hand Pose Estimation Based on Finger-Point Reinforcement and Multi-Level Feature Fusion[J]. Acta Optica Sinica, 2022, 42(19): 1915001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Jan. 19, 2022

Accepted: Apr. 16, 2022

Published Online: Oct. 18, 2022

The Author Email: Da Feipeng (dafp@seu.edu.cn)

DOI:10.3788/AOS202242.1915001

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology