Real-Time Multi-Person Video-Based Pose Estimation

Yan Fenting; Wang Peng; Lü Zhigang; Ding Zhe; Qiao Mengyu

doi:10.3788/LOP57.021006

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 2 >Page 21006 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 2, 21006 (2020)

Real-Time Multi-Person Video-Based Pose Estimation

Yan Fenting, Wang Peng^*, Lü Zhigang, Ding Zhe, and Qiao Mengyu

Author Affiliations

School of Electronics and Information Engineering, Xi''an Technological University, Xi''an, Shaanxi 710021, China

show less

DOI: 10.3788/LOP57.021006 Cite this Article Set citation alerts

Yan Fenting, Wang Peng, Lü Zhigang, Ding Zhe, Qiao Mengyu. Real-Time Multi-Person Video-Based Pose Estimation[J]. Laser & Optoelectronics Progress, 2020, 57(2): 21006 Copy Citation Text

show less

Fig. 1. Structural diagram of symmetric space transformation network

Download full size

Fig. 2. Real-time multi-person pose estimation model

Download full size

Fig. 3. Pose estimation network model

Download full size

Fig. 4. [in Chinese]

Download full size

Fig. 4. Comparison of results in different scenarios for each model. (a)-(d) Scale change; (e)-(h) dense population; (i)-(l) occlusion; (m)-(p) complex pose

Download full size

Model	Framework	Programminglanguage
CMU-Pose^[10]	Caffe	Python3.6.2
MaskR-CNN^[11]	TensorFlow1.3.0+Keras2.2.6	Python3.6.2
RMPE^[14]	Pytorch0.4.0	Python3.6.2
Proposed model	Pytorch0.4.0	Python3.6.2

Table 1. Algorithm model and environment configuration

Model	AP	AP_@0.5	AP_@0.75	AP_m	AP_l
CMU-Pose	61.8	84.9	67.5	57.1	68.2
Mask R-CNN	63.1	87.3	68.7	57.8	71.4
RMPE	72.3	89.2	79.1	68.0	78.6
Proposed model	74.1	92.5	80.5	70.6	79.5

Table 2. Comparison of performance of each pose estimation model

Model	Data set	Running speed /(frame·s^-1)	Parametersize /MB	Calculatedamount /10⁹
YOLOv3^[15]	MS COCO	51	237	65.86
Proposed model	MS COCO	64	195	44.32

Table 3. Comparison of parameters of each human detection algorithm

Input	AP	AP_@0.5	AP_@0.75	AP_m	AP_l	AR	AR_@0.5	AR_@0.75	AR_m	AR_l
256 pixel×192 pixel	71.2	91.4	78.3	68.5	75.2	74.3	92.2	80.9	71.3	78.9
384 pixel×288 pixel	74.1	92.5	80.5	70.6	79.5	76.8	93.2	82.5	73.0	82.6

Table 4. AP-AR values of model under different inputs

Yan Fenting, Wang Peng, Lü Zhigang, Ding Zhe, Qiao Mengyu. Real-Time Multi-Person Video-Based Pose Estimation[J]. Laser & Optoelectronics Progress, 2020, 57(2): 21006

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information