3D object detection in automatic driving scene clustering
DOI:
CSTR:
Author:
Affiliation:

1.Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China; 2.Key Laboratory of Electromagnetic Radiation and Sensing Technology, Chinese Academy of Sciences, Beijing 100190, China; University of Chinese Academy of Sciences, School of Electronic, Electrical and Communication Engineering, Beijing 100049, China

Clc Number:

TN919.8

Fund Project:

  • Article
  • |
  • Figures
  • |
  • Metrics
  • |
  • Reference
  • |
  • Related
  • |
  • Cited by
  • |
  • Materials
  • |
  • Comments
    Abstract:

    KITTI is a large data set fused with multiple sensors in automatic driving scene, its data acquisition platform includes two gray-scale cameras, two color cameras, a velodyne 64 line lidar, four optical lenses and a GPS navigation system. KITTI 3D Object Detection Evaluation can verify the accuracy and effectiveness of various 3D object detection algorithms. It is the most important data set in the field of autonomous driving. The focus of this article is the data reconstruction and data cleaning of the KITTI data set: first, use the RANSAC algorithm to remove the ground from each frame of lidar data in the KITTI data set, and use the DBSCAN algorithm to cluster the targets on the ground, and then according to the label The file uses the nearest neighbor search to assign tags to each target category to complete the data reconstruction. Based on this, the data is resampled to balance the categories to complete the data cleaning. For the reconstructed and cleaned KITTI data, the PointNet algorithm is used to complete the classification task, and the accuracy rate is as high as 95.13%. Finally, the overall framework of 3D target detection and evaluation on the KITTI data set is completed. The results show that the quality of the reconstructed and cleaned new data set is high, the classification algorithm is robust, and the 3D target detection process is clear and complete.

    Reference
    Related
    Cited by
Get Citation
Share
Article Metrics
  • Abstract:
  • PDF:
  • HTML:
  • Cited by:
History
  • Received:
  • Revised:
  • Adopted:
  • Online: October 18,2024
  • Published:
Article QR Code