Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Kalman-Based Scene Flow Estimation for Point Cloud Densification and 3D Object Detection in Dynamic Scenes.

Junzhe Ding1, Jin Zhang1, Luqin Ye1

  • 1School of Rail Transportation, Soochow University, Suzhou 215500, China.

Sensors (Basel, Switzerland)
|February 10, 2024
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Human Fall Detection Based on Body Posture Spatio-Temporal Evolution.

Sensors (Basel, Switzerland)ยท2020
See all related articles

This study introduces a Kalman-based method to improve 3D object detection by densifying point clouds in dynamic scenes. The approach corrects localization errors, enhancing shape completion and detection accuracy for LiDAR-only systems.

Area of Science:

  • Computer Vision
  • Robotics
  • 3D Perception

Background:

  • Point cloud densification is vital for 3D environment understanding and tasks like object detection.
  • Existing registration methods fail with dynamic targets due to incomplete and deformed point clouds.
  • Localization errors in scene flow estimation hinder accurate 3D perception.

Purpose of the Study:

  • To develop a novel Kalman-based scene flow estimation method for point cloud densification.
  • To enhance 3D object detection accuracy in dynamic scenes.
  • To address and correct localization errors in dynamic scene flow estimation.

Main Methods:

  • A Kalman filter is integrated to correct dynamic target positions during scene flow estimation.
  • The method estimates long-sequence scene flow, mitigating cumulative localization errors.
Keywords:
3D object detectionKalman filterpoint cloud densificationscene flow estimation

Related Experiment Videos

  • Point cloud densification is achieved through accurate scene flow estimation.

    • Main Results:

    • Significantly improved performance for LiDAR-only 3D object detectors.
    • Enhanced accuracy and precision in shape completion for dynamic targets.
    • Superior results compared to baseline methods on the KITTI 3D tracking dataset.

    Conclusions:

    • The proposed Kalman-based method effectively handles dynamic scenes for point cloud densification.
    • It overcomes limitations of existing registration-based approaches.
    • This technique offers a robust solution for accurate 3D object detection in complex, dynamic environments.