monocular visual-imu odometry a comparative evaluation of

16
The University of Manchester Research Monocular Visual-IMU Odometry: A Comparative Evaluation of the Detector-Descriptor Based Methods DOI: 10.1007/978-3-319-46604-0_6 Document Version Accepted author manuscript Link to publication record in Manchester Research Explorer Citation for published version (APA): Dong, X., Dong, X., & Dong, J. (2016). Monocular Visual-IMU Odometry: A Comparative Evaluation of the Detector-Descriptor Based Methods. In G. Hua, & H. Jegou (Eds.), Computer vision - ECCV 2016 workshops : Amsterdam, The Netherlands, October 8-10 and 15-16, 2016, proceedings (Vol. Part 1, pp. 81-95). (Lecture Notes in Computer Science; Vol. 9913). Springer Nature. https://doi.org/10.1007/978-3-319-46604-0_6 Published in: Computer vision - ECCV 2016 workshops Citing this paper Please note that where the full-text provided on Manchester Research Explorer is the Author Accepted Manuscript or Proof version this may differ from the final Published version. If citing, it is advised that you check and use the publisher's definitive version. General rights Copyright and moral rights for the publications made accessible in the Research Explorer are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. Takedown policy If you believe that this document breaches copyright please refer to the University of Manchester’s Takedown Procedures [http://man.ac.uk/04Y6Bo] or contact [email protected] providing relevant details, so we can investigate your claim. Download date:21. Feb. 2022

Upload: others

Post on 21-Feb-2022

10 views

Category:

Documents


0 download

TRANSCRIPT

The University of Manchester Research

Monocular Visual-IMU Odometry: A ComparativeEvaluation of the Detector-Descriptor Based MethodsDOI:10.1007/978-3-319-46604-0_6

Document VersionAccepted author manuscript

Link to publication record in Manchester Research Explorer

Citation for published version (APA):Dong, X., Dong, X., & Dong, J. (2016). Monocular Visual-IMU Odometry: A Comparative Evaluation of theDetector-Descriptor Based Methods. In G. Hua, & H. Jegou (Eds.), Computer vision - ECCV 2016 workshops :Amsterdam, The Netherlands, October 8-10 and 15-16, 2016, proceedings (Vol. Part 1, pp. 81-95). (Lecture Notesin Computer Science; Vol. 9913). Springer Nature. https://doi.org/10.1007/978-3-319-46604-0_6Published in: Computer vision - ECCV 2016 workshops

Citing this paperPlease note that where the full-text provided on Manchester Research Explorer is the Author Accepted Manuscriptor Proof version this may differ from the final Published version. If citing, it is advised that you check and use thepublisher's definitive version.

General rightsCopyright and moral rights for the publications made accessible in the Research Explorer are retained by theauthors and/or other copyright owners and it is a condition of accessing publications that users recognise andabide by the legal requirements associated with these rights.

Takedown policyIf you believe that this document breaches copyright please refer to the University of Manchester’s TakedownProcedures [http://man.ac.uk/04Y6Bo] or contact [email protected] providingrelevant details, so we can investigate your claim.

Download date:21. Feb. 2022

*Corresponding author: Xinghui Dong. Email: [email protected].

Supplementary Material for ECCV 2016 Workshop

Paper: Monocular Visual-IMU Odometry: A

Comparative Evaluation of the Detector-Descriptor

Based Methods

Xingshuai Dong a, Xinghui Dong

b*, and Junyu Dong

a

a

Ocean University of China, Qingdao, 266071, China b Centre for Imaging Sciences, University of Manchester, Manchester, M13 9PT, UK

Abstract. In Sections 1 to 5, we provide the figures of all the trajectories that

are obtained in the five experiments introduced in the main paper. Each of these

sections corresponds to a subsection of Section 5 in the main paper, i.e., an in-

dividual experiment. It should be noted that the figures shown here is not neces-

sary for understanding the content of the main paper. In addition, we give the

details of the five salient point detectors and eight local descriptors tested in this

study and the configurations of five routes in Sections 6 and 7 respectively.

1 Route 1: Straight Line

In this section, five figures are displayed in terms of five different salient point detec-

tors respectively. In each figure, the ground-truth trajectory (in black) and the esti-

mated trajectories (in different colors) obtained using the IMU method and eight dif-

ferent local descriptors on the Straight Line route are shown together.

Fig. 1. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Blob&Corner salient point detector and different local descriptors on the

Straight Line route (express way, ≈780m, ≈60km/h). (Map source: GoogleEarth).

8.445 8.446 8.447 8.448 8.449 8.45 8.451 8.452 8.453 8.454 8.455

49.001

49.002

49.003

49.004

49.005

49.006

49.007

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO BLOB&CORNER+ICIMGPVIO BLOB&CORNER+IMGPVIO BLOB&CORNER+LIOPVIO BLOB&CORNER+LMVIO BLOB&CORNER+LSSDVIO BLOB&CORNER+SIFTVIO BLOB&CORNER+SURFVIO BLOB&CORNER+HOG

8.445 8.4455 8.446 8.4465 8.44749.003

49.0035

49.004

Fig. 2. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the DoG salient point detector and different local descriptors on the Straight

Line route (express way, ≈780m, ≈60km/h). (Map source: GoogleEarth).

Fig. 3. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the FAST salient point detector and different local descriptors on the Straight

Line route (express way, ≈780m, ≈60km/h). (Map source: GoogleEarth).

8.445 8.446 8.447 8.448 8.449 8.45 8.451 8.452 8.453 8.454 8.455

49.001

49.002

49.003

49.004

49.005

49.006

49.007

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO DOG+ICIMGPVIO DOG+IMGPVIO DOG+LIOPVIO DOG+LMVIO DOG+LSSDVIO DOG+SIFTVIO DOG+SURFVIO DOG+HOG

8.4445 8.445 8.4455 8.44649.003

49.0035

49.004

8.445 8.446 8.447 8.448 8.449 8.45 8.451 8.452 8.453 8.454 8.455

49.001

49.002

49.003

49.004

49.005

49.006

49.007

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO FAST+ICIMGPVIO FAST+IMGPVIO FAST+LIOPVIO FAST+LMVIO FAST+LSSDVIO FAST+SIFTVIO FAST+SURFVIO FAST+HOG

8.444 8.4445 8.445 8.4455

49.0035

49.004

49.0045

Fig. 4. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Fast Hessian salient point detector and different local descriptors on the

Straight Line route (express way, ≈780m, ≈60km/h). (Map source: GoogleEarth).

Fig. 5. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Harris Laplace salient point detector and different local descriptors on the

Straight Line route (express way, ≈780m, ≈60km/h). (Map source: GoogleEarth).

2 Route 2: Quarter Turn

This section shows five figures in terms of five different salient point detectors re-

spectively. In each figure, the ground-truth trajectory (in black) and the estimated

trajectories (in different colors) obtained using the IMU method and eight different

local descriptors on the Quarter Turn route are shown together.

8.445 8.446 8.447 8.448 8.449 8.45 8.451 8.452 8.453 8.454 8.455

49.001

49.002

49.003

49.004

49.005

49.006

49.007

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO FASTHESSIAN+ICIMGPVIO FASTHESSIAN+IMGPVIO FASTHESSIAN+LIOPVIO FASTHESSIAN+LMVIO FASTHESSIAN+LSSDVIO FASTHESSIAN+SIFTVIO FASTHESSIAN+SURFVIO FASTHESSIAN+HOG

8.444 8.4445 8.445 8.4455 8.446

49.0035

49.004

49.0045

8.445 8.446 8.447 8.448 8.449 8.45 8.451 8.452 8.453 8.454 8.455

49.001

49.002

49.003

49.004

49.005

49.006

49.007

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO HARRISLAPLACE+ICIMGPVIO HARRISLAPLACE+IMGPVIO HARRISLAPLACE+LIOPVIO HARRISLAPLACE+LMVIO HARRISLAPLACE+LSSDVIO HARRISLAPLACE+SIFTVIO HARRISLAPLACE+SURFVIO HARRISLAPLACE+HOG

8.4445 8.445 8.4455 8.446 8.4465

49.003249.003449.003649.0038

49.00449.0042

Fig. 6. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Blob&Corner salient point detector and different local descriptors on the

Quarter Turn route (urban road, ≈330m, ≈28km/h). (Map source: GoogleEarth).

Fig. 7. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the DoG salient point detector and different local descriptors on the Quarter

Turn route (urban road, ≈330m, ≈28km/h). (Map source: GoogleEarth).

8.4375 8.438 8.4385 8.439 8.4395 8.44 8.4405 8.441

49.009

49.0095

49.01

49.0105

49.011

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO BLOB&CORNER+ICIMGPVIO BLOB&CORNER+IMGPVIO BLOB&CORNER+LIOPVIO BLOB&CORNER+LMVIO BLOB&CORNER+LSSDVIO BLOB&CORNER+SIFTVIO BLOB&CORNER+SURFVIO BLOB&CORNER+HOG

8.4402 8.4404 8.4406 8.4408

49.0098

49.0099

49.01

49.0101

49.0102

8.4375 8.438 8.4385 8.439 8.4395 8.44 8.4405 8.441

49.009

49.0095

49.01

49.0105

49.011

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO DOG+ICIMGPVIO DOG+IMGPVIO DOG+LIOPVIO DOG+LMVIO DOG+LSSDVIO DOG+SIFTVIO DOG+SURFVIO DOG+HOG

8.4402 8.4404 8.4406 8.440849.0097

49.0098

49.0099

49.01

49.0101

Fig. 8. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the FAST salient point detector and different local descriptors on the Quarter

Turn route (urban road, ≈330m, ≈28km/h). (Map source: GoogleEarth).

Fig. 9. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Fast Hessian salient point detector and different local descriptors on the

Quarter Turn route (urban road, ≈330m, ≈28km/h). (Map source: GoogleEarth).

8.437 8.4375 8.438 8.4385 8.439 8.4395 8.44 8.4405

49.009

49.0095

49.01

49.0105

49.011

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO FAST+ICIMGPVIO FAST+IMGPVIO FAST+LIOPVIO FAST+LMVIO FAST+LSSDVIO FAST+SIFTVIO FAST+SURFVIO FAST+HOG

8.4402 8.4404 8.4406 8.4408

49.0097

49.0098

49.0099

49.01

49.0101

8.4375 8.438 8.4385 8.439 8.4395 8.44 8.4405 8.441

49.009

49.0095

49.01

49.0105

49.011

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO FASTHESSIAN+ICIMGPVIO FASTHESSIAN+IMGPVIO FASTHESSIAN+LIOPVIO FASTHESSIAN+LMVIO FASTHESSIAN+LSSDVIO FASTHESSIAN+SIFTVIO FASTHESSIAN+SURFVIO FASTHESSIAN+HOG

8.44 8.4402 8.4404 8.4406

49.0097

49.0098

49.0099

49.01

49.0101

Fig. 10. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Harris Laplace salient point detector and different local descriptors on the

Quarter Turn route (urban road, ≈330m, ≈28km/h). (Map source: GoogleEarth).

3 Route 3: Multiple Quarter Turns

This section shows five figures in terms of five different salient point detectors re-

spectively. In each figure, the ground-truth trajectory (in black) and the estimated

trajectories (in different colors) obtained using the IMU method and eight different

local descriptors on the Multiple Quarter Turns route are shown together.

Fig. 11. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Blob&Corner salient point detector and different local descriptors on the

Multiple Quarter Turns route (residential area, ≈ 960m, ≈ 26km/h). (Map source:

GoogleEarth).

8.4375 8.438 8.4385 8.439 8.4395 8.44 8.4405 8.441

49.009

49.0095

49.01

49.0105

49.011

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO HARRISLAPLACE+ICIMGPVIO HARRISLAPLACE+IMGPVIO HARRISLAPLACE+LIOPVIO HARRISLAPLACE+LMVIO HARRISLAPLACE+LSSDVIO HARRISLAPLACE+SIFTVIO HARRISLAPLACE+SURFVIO HARRISLAPLACE+HOG

8.4404 8.4406 8.4408 8.441

49.0098

49.0099

49.01

49.0101

49.0102

8.39 8.391 8.392 8.393 8.394 8.395 8.396

48.9825

48.983

48.9835

48.984

48.9845

48.985

48.9855

48.986

48.9865

48.987

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO BLOB&CORNER+ICIMGPVIO BLOB&CORNER+IMGPVIO BLOB&CORNER+LIOPVIO BLOB&CORNER+LMVIO BLOB&CORNER+LSSDVIO BLOB&CORNER+SIFTVIO BLOB&CORNER+SURFVIO BLOB&CORNER+HOG

8.391 8.3915 8.392

48.983

48.9832

48.9834

48.9836

Fig. 12. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the DoG salient point detector and different local descriptors on the Multiple

Quarter Turns route (residential area, ≈960m, ≈26km/h). (Map source: GoogleEarth).

Fig. 13. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the FAST salient point detector and different local descriptors on the Multiple

Quarter Turns route (residential area, ≈960m, ≈26km/h). (Map source: GoogleEarth).

8.389 8.39 8.391 8.392 8.393 8.394 8.395 8.396

48.9825

48.983

48.9835

48.984

48.9845

48.985

48.9855

48.986

48.9865

48.987

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO DOG+ICIMGPVIO DOG+IMGPVIO DOG+LIOPVIO DOG+LMVIO DOG+LSSDVIO DOG+SIFTVIO DOG+SURFVIO DOG+HOG

8.391 8.3915 8.392

48.9828

48.983

48.9832

48.9834

48.9836

8.39 8.391 8.392 8.393 8.394 8.395 8.396

48.9825

48.983

48.9835

48.984

48.9845

48.985

48.9855

48.986

48.9865

48.987

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO FAST+ICIMGPVIO FAST+IMGPVIO FAST+LIOPVIO FAST+LMVIO FAST+LSSDVIO FAST+SIFTVIO FAST+SURFVIO FAST+HOG

8.391 8.3915 8.392

48.983

48.9832

48.9834

48.9836

Fig. 14. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Fast Hessian salient point detector and different local descriptors on the

Multiple Quarter Turns route (residential area, ≈ 960m, ≈ 26km/h). (Map source:

GoogleEarth).

Fig. 15. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Harris Laplace salient point detector and different local descriptors on the

Multiple Quarter Turns route (residential area, ≈ 960m, ≈ 26km/h). (Map source:

GoogleEarth).

4 Route 4: Multiple Curved Turns

In this section, five figures are displayed in terms of five different salient point detec-

tors respectively. In each figure, the ground-truth trajectory (in black) and the esti-

8.389 8.39 8.391 8.392 8.393 8.394 8.395 8.39648.9825

48.983

48.9835

48.984

48.9845

48.985

48.9855

48.986

48.9865

48.987

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO FASTHESSIAN+ICIMGPVIO FASTHESSIAN+IMGPVIO FASTHESSIAN+LIOPVIO FASTHESSIAN+LMVIO FASTHESSIAN+LSSDVIO FASTHESSIAN+SIFTVIO FASTHESSIAN+SURFVIO FASTHESSIAN+HOG

8.3905 8.391 8.3915

48.983

48.9832

48.9834

48.9836

8.39 8.391 8.392 8.393 8.394 8.395 8.396

48.983

48.9835

48.984

48.9845

48.985

48.9855

48.986

48.9865

48.987

48.9875

Longitude

Latit

ude

Motion Trajectory

Ground−TruthIMUVIO HARRISLAPLACE+ICIMGPVIO HARRISLAPLACE+IMGPVIO HARRISLAPLACE+LIOPVIO HARRISLAPLACE+LMVIO HARRISLAPLACE+LSSDVIO HARRISLAPLACE+SIFTVIO HARRISLAPLACE+SURFVIO HARRISLAPLACE+HOG

8.3905 8.391 8.3915

48.983

48.9832

48.9834

48.9836

mated trajectories (in different colors) obtained using the IMU method and eight dif-

ferent local descriptors on the Multiple Curved Turns route are shown together.

Fig. 16. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Blob&Corner salient point detector and different local descriptors on the

Multiple Curved Turns route (residential area, ≈ 1050m, ≈ 40km/h). (Map source:

GoogleEarth).

Fig. 17. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the DoG salient point detector and different local descriptors on the Multiple

Curved Turns route (residential area, ≈1050m, ≈40km/h). (Map source: GoogleEarth).

8.47 8.472 8.474 8.476 8.478 8.48 8.482

48.984

48.985

48.986

48.987

48.988

48.989

48.99

48.991

48.992

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO BLOB&CORNER+ICIMGP

VIO BLOB&CORNER+IMGP

VIO BLOB&CORNER+LIOP

VIO BLOB&CORNER+LM

VIO BLOB&CORNER+LSSD

VIO BLOB&CORNER+SIFT

VIO BLOB&CORNER+SURF

VIO BLOB&CORNER+HOG

8.4735 8.474 8.4745 8.475 8.4755

48.986

48.9865

48.987

8.4788.47858.4798.4795 8.4848.987

48.9875

48.988

48.9885

8.47 8.472 8.474 8.476 8.478 8.48 8.48248.983

48.984

48.985

48.986

48.987

48.988

48.989

48.99

48.991

48.992

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO DOG+ICIMGP

VIO DOG+IMGP

VIO DOG+LIOP

VIO DOG+LM

VIO DOG+LSSD

VIO DOG+SIFT

VIO DOG+SURF

VIO DOG+HOG

8.4735 8.474 8.4745 8.475 8.4755

48.986

48.9865

48.987

8.478 8.479 8.48

48.9875

48.988

48.9885

48.989

Fig. 18. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the FAST salient point detector and different local descriptors on the Multiple

Curved Turns route (residential area, ≈1050m, ≈40km/h). (Map source: GoogleEarth).

Fig. 19. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Fast Hessian salient point detector and different local descriptors on the

Multiple Curved Turns route (residential area, ≈ 1050m, ≈ 40km/h). (Map source:

GoogleEarth).

8.47 8.472 8.474 8.476 8.478 8.48 8.482

48.984

48.985

48.986

48.987

48.988

48.989

48.99

48.991

48.992

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO FAST+ICIMGP

VIO FAST+IMGP

VIO FAST+LIOP

VIO FAST+LM

VIO FAST+LSSD

VIO FAST+SIFT

VIO FAST+SURF

VIO FAST+HOG

8.4735 8.474 8.4745 8.475 8.475548.9855

48.986

48.9865

48.987

8.4775 8.478 8.4785 8.479 8.4795 8.48

48.9875

48.988

48.9885

8.47 8.472 8.474 8.476 8.478 8.48 8.482

48.983

48.984

48.985

48.986

48.987

48.988

48.989

48.99

48.991

48.992

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO FASTHESSIAN+ICIMGP

VIO FASTHESSIAN+IMGP

VIO FASTHESSIAN+LIOP

VIO FASTHESSIAN+LM

VIO FASTHESSIAN+LSSD

VIO FASTHESSIAN+SIFT

VIO FASTHESSIAN+SURF

VIO FASTHESSIAN+HOG

8.473 8.474 8.47548.9855

48.986

48.9865

48.987

8.4788.47858.4798.4795 8.48

48.9875

48.988

48.9885

48.989

Fig. 20. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Harris Laplace salient point detector and different local descriptors on the

Multiple Curved Turns route (residential area, ≈ 1050m, ≈ 40km/h). (Map source:

GoogleEarth).

5 Route 5: Loop Line

This section displays five figures in terms of five different salient point detectors re-

spectively. In each figure, the ground-truth trajectory (in black) and the estimated

trajectories (in different colors) obtained using the IMU method and eight different

local descriptors on the Loop Line route are shown together.

Fig. 21. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Blob&Corner salient point detector and different local descriptors on the

Loop Line route (residential area, ≈930m, ≈38km/h). (Map source: GoogleEarth).

8.47 8.472 8.474 8.476 8.478 8.48 8.482

48.984

48.985

48.986

48.987

48.988

48.989

48.99

48.991

48.992

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO HARRISLAPLACE+ICIMGP

VIO HARRISLAPLACE+IMGP

VIO HARRISLAPLACE+LIOP

VIO HARRISLAPLACE+LM

VIO HARRISLAPLACE+LSSD

VIO HARRISLAPLACE+SIFT

VIO HARRISLAPLACE+SURF

VIO HARRISLAPLACE+HOG

8.4735 8.474 8.4745 8.475 8.4755

48.9855

48.986

48.9865

48.987

8.4785 8.479 8.4795 8.48 8.4805

48.9875

48.988

48.9885

48.989

8.393 8.394 8.395 8.396 8.397 8.398 8.399 8.4

49.0515

49.052

49.0525

49.053

49.0535

49.054

49.0545

49.055

49.0555

49.056

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO BLOB&CORNER+ICIMGP

VIO BLOB&CORNER+IMGP

VIO BLOB&CORNER+LIOP

VIO BLOB&CORNER+LM

VIO BLOB&CORNER+LSSD

VIO BLOB&CORNER+SIFT

VIO BLOB&CORNER+SURF

VIO BLOB&CORNER+HOG

8.39268.3928 8.393 8.393249.0529

49.053

49.0531

49.0532

49.0533

8.399 8.39928.39948.399649.0533

49.0534

49.0535

49.0536

49.0537

Fig. 22. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the DoG salient point detector and different local descriptors on the Loop Line

route (residential area, ≈930m, ≈38km/h). (Map source: GoogleEarth).

Fig. 23. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the FAST salient point detector and different local descriptors on the Loop

Line route (residential area, ≈930m, ≈38km/h). (Map source: GoogleEarth).

8.393 8.394 8.395 8.396 8.397 8.398 8.399 8.4

49.0515

49.052

49.0525

49.053

49.0535

49.054

49.0545

49.055

49.0555

49.056

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO DOG+ICIMGP

VIO DOG+IMGP

VIO DOG+LIOP

VIO DOG+LM

VIO DOG+LSSD

VIO DOG+SIFT

VIO DOG+SURF

VIO DOG+HOG

8.393 8.3935 8.39449.0528

49.053

49.0532

49.0534

49.0536

8.3985 8.399 8.3995

49.0532

49.0534

49.0536

49.0538

49.054

8.393 8.394 8.395 8.396 8.397 8.398 8.399 8.4

49.0515

49.052

49.0525

49.053

49.0535

49.054

49.0545

49.055

49.0555

49.056

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO FAST+ICIMGP

VIO FAST+IMGP

VIO FAST+LIOP

VIO FAST+LM

VIO FAST+LSSD

VIO FAST+SIFT

VIO FAST+SURF

VIO FAST+HOG

8.39268.3928 8.393 8.3932

49.053

49.0531

49.0532

49.0533

49.0534

8.399 8.3992 8.3994 8.3996

49.0533

49.0534

49.0535

49.0536

49.0537

Fig. 24. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Fast Hessian salient point detector and different local descriptors on the

Loop Line route (residential area, ≈930m, ≈38km/h). (Map source: GoogleEarth).

Fig. 25. The ground-truth motion trajectory and the trajectories obtained using IMU and the

combinations of the Harris Laplace salient point detector and different local descriptors on the

Loop Line route (residential area, ≈930m, ≈38km/h). (Map source: GoogleEarth).

6 Parameter Values of the Salient Point Detectors and Local

Descriptors

In this section, we list the parameters used for the five salient point detectors and

eight local descriptors in detail.

8.393 8.394 8.395 8.396 8.397 8.398 8.399 8.4

49.0515

49.052

49.0525

49.053

49.0535

49.054

49.0545

49.055

49.0555

49.056

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO FASTHESSIAN+ICIMGP

VIO FASTHESSIAN+IMGP

VIO FASTHESSIAN+LIOP

VIO FASTHESSIAN+LM

VIO FASTHESSIAN+LSSD

VIO FASTHESSIAN+SIFT

VIO FASTHESSIAN+SURF

VIO FASTHESSIAN+HOG

8.3925 8.393 8.393549.0526

49.0528

49.053

49.0532

49.0534

8.399 8.3995 8.4

49.0534

49.0536

49.0538

49.054

8.393 8.394 8.395 8.396 8.397 8.398 8.399 8.4

49.0515

49.052

49.0525

49.053

49.0535

49.054

49.0545

49.055

49.0555

49.056

Longitude

Latit

ude

Motion Trajectory

Ground−Truth

IMU

VIO HARRISLAPLACE+ICIMGP

VIO HARRISLAPLACE+IMGP

VIO HARRISLAPLACE+LIOP

VIO HARRISLAPLACE+LM

VIO HARRISLAPLACE+LSSD

VIO HARRISLAPLACE+SIFT

VIO HARRISLAPLACE+SURF

VIO HARRISLAPLACE+HOG

8.3928 8.393 8.39328.3934

49.053

49.0531

49.0532

49.0533

49.0534

8.399 8.39928.39948.3996

49.0534

49.0535

49.0536

49.0537

49.0538

6.1 Parameter Values of the Salient Point Detectors

Detector Parameter Value

Blob&Corner Window Size 13×13

DoG Octaves 4

Levels Per Octave 3

�� 1.6

Fast Hessian Octaves 4

Threshold � 0.2

Sampling Step � 2

FAST Threshold � 30

Harris-Laplace Peak Threshold 0.000002

Edge Threshold 10

6.2 Parameter Values of the Local Descriptors

Detector Parameter Value

HOG Block Size 15×15

Cell Size 5×5

Number of Orientations 9

IMGP Patch Size 11×11

ICIMGP Number of Channels 8

Patch Size 11×11

LM �� √2, 2, 2√2

LIOP Number of Spatial Bins 6

Number of Neighbors 4

Sampling Radius 6

LSSD Patch Size 5×5

Region Radius 40

� 12

�� 3

SIFT Sub-window Size 4×4

Number of Bins Per Window 3

SURF Window Size 4×4

7 Configurations of the Five Routes

We describe the details of each route used in this study in terms of the information

provided with the KITTI dataset in the table below.

Route

No.

Sequence No. in the KITTI

Dataset

No. of the Start

Point Image (The

Left Color Cam-

era Image)

No. of the End

Point Image (The

Left Color Cam-

era Image)

1 2011_09_26_drive_0101_sync 0000000361 0000000816

2 2011_09_26_drive_0009_sync

0000000000

0000000405

3 2011_10_03_drive_0027_sync 0000000301

0000001610

4 2011_10_03_drive_0034_sync 0000000000

0000000899

5 2011_09_30_drive_0020_sync 0000000000

0000000834