Object Pose Estimation Method Based on Keypoint Distance Network

Meng Xia; Hongzhi Du; Jiarui Lin; Yanbiao Sun; Jigui Zhu

doi:10.3788/LOP223015

Journals >Laser & Optoelectronics Progress >Volume 60 >Issue 16 >Page 1615008 > Article

Laser & Optoelectronics Progress
Vol. 60, Issue 16, 1615008 (2023)

Object Pose Estimation Method Based on Keypoint Distance Network

Meng Xia, Hongzhi Du, Jiarui Lin, Yanbiao Sun^*, and Jigui Zhu

Author Affiliations

State Key Laboratory of Precision Measurement Technology and Instrument, Tianjin University, Tianjin 300072, China

show less

DOI: 10.3788/LOP223015 Cite this Article Set citation alerts

Meng Xia, Hongzhi Du, Jiarui Lin, Yanbiao Sun, Jigui Zhu. Object Pose Estimation Method Based on Keypoint Distance Network[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1615008 Copy Citation Text

show less

Fig. 1. Overview of pose estimation

Download full size

Fig. 2. Backbone network module

Download full size

Fig. 3. Feature fusion structure

Download full size

Fig. 4. Confidence network

Download full size

Fig. 5. Flow chart of reasoning calculation

Download full size

Fig. 6. Visualization results. (a) LineMOD dataset; (b) YCB-Video dataset

Download full size

Fig. 7. Comparison of pose estimation results

Download full size

Object	PoseCNN	PVNet	DenseFusion	G2L-Net	Proposed method
mean	88.6	86.3	94.3	98.7	99.8
ape	77.0	43.6	92.3	96.8	98.7
benchvise	97.5	99.9	93.2	96.1	100.0
camera	93.5	86.9	94.4	98.2	99.9
can	96.5	95.5	93.1	98.0	100.0
cat	82.1	79.3	96.5	99.2	99.9
driller	95.0	96.4	87.0	99.8	100.0
duck	77.7	52.6	92.3	97.7	98.6
eggbox	97.1	99.2	99.8	100.0	100.0
glue	99.4	95.7	100.0	100.0	99.9
holepuncher	52.8	82.0	92.1	99.0	100.0
iron	98.3	98.9	97.0	99.3	100.0
lamp	97.5	99.3	95.3	99.5	100.0
phone	87.7	92.4	92.8	98.9	100.0

Table 1. Test results on LineMOD dataset [10%d-ADD(S)]

Object	PoseCNN		DenseFusion		GDR-Net		Proposed method
Object	ADD-S	ADD（S）	ADD-S	ADD（S）	ADD-S	ADD（S）	ADD-S	ADD（S）
mean	75.9	59.9	91.2	82.9	91.6	84.3	94.1	90.1
02 master chef can	83.9	50.2	95.3	70.7	96.3	65.2	95.1	77.7
03 cracker box	76.9	53.1	92.5	86.9	97.0	88.8	93.9	89.4
04 sugar box	84.2	68.4	95.1	90.8	98.9	95.0	96.5	94.4
05 tomato soup can	81.0	66.2	93.8	84.7	96.5	91.9	94.9	89.2
06 mustard bottle	90.4	81.0	95.8	90.9	100.0	92.8	96.8	94.7
07 tuna fish can	88.0	70.7	95.7	79.6	99.4	94.2	95.0	90.3
08 pudding box	79.1	62.7	94.3	89.3	64.6	44.7	93.9	87.4
09 gelatin box	87.2	75.2	97.2	95.8	97.1	92.5	97.0	94.8
10 potted meat can	78.5	59.5	89.3	79.6	86.0	80.2	89.9	81.8
11 banana	86.0	72.3	90.0	76.7	96.3	85.8	95.2	91.1
19 pitcher base	77.0	53.3	93.6	87.1	99.9	98.5	96.6	94.9
21 bleach cleanser	71.6	50.3	94.4	87.5	94.2	84.3	95.0	90.6
24 bowl	69.6	69.6	86.0	86.0	85.7	85.7	88.1	88.1
25 mug	78.2	58.5	95.3	83.8	99.6	94.0	96.8	92.5
35 power drill	72.7	55.3	92.1	83.7	97.5	90.1	95.6	92.7
36 wood block	64.3	64.3	89.5	89.5	82.5	82.5	90.5	90.5
37 scissors	56.9	35.8	90.1	77.4	63.8	49.5	93.6	89.9
40 large marker	71.7	58.3	95.1	89.1	88.0	76.1	95.0	85.3
51 large clamp	50.2	50.2	71.5	71.5	89.3	89.3	93.3	93.3
52 extra large clamp	44.1	44.1	70.2	70.2	93.5	93.5	88.2	88.2
61 foam brick	88.0	88.0	92.2	92.2	96.9	96.9	95.8	95.8

Table 2. Test results on YCB-Video dataset [AUC-ADD(S)]

Dataset	ResNet	Proposed module
LineMOD	99.6	99.8
YCB-Video，ADD-S	92.1	94.1
YCB-Video，ADD（S）	86.5	90.1

Table 3. Accuracy comparison of different backbone network modules

Module	ResNet	Proposed module
Parameters /MB	33.7	15.4

Table 4. Feature network parameters of different backbone network modules

Dataset	Without data processing	With data processing
LineMOD	96.1	96.6
YCB-Video，ADD-S	93.5	94.1
YCB-Video，ADD（S）	88.3	90.1

Table 5. Comparison of results with or without data processing

Meng Xia, Hongzhi Du, Jiarui Lin, Yanbiao Sun, Jigui Zhu. Object Pose Estimation Method Based on Keypoint Distance Network[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1615008

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享