Laser Point Cloud-Image Fusion Technology for Intelligent Driving Vehicles Based on Semi-Supervised Learning Algorithm

In autonomous vehicle, the traditional method of manually labeling laser point cloud data is not only expensive, but also complex in process. To overcome these challenges, a laser point cloud-image fusion technology based on semi-supervised learning algorithms is proposed to improve the efficiency a...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:IEEE access Ročník 12; s. 132664 - 132676
Hlavní autori: Lan, Jianping, Dong, Xiujuan
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Piscataway IEEE 2024
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Predmet:
3D object detection
Accuracy
Algorithms
Calibration
camera-LiDAR system
Cameras
Computer vision
image fusion
laser point cloud
Laser radar
Lasers
Machine learning
Object detection
Object recognition
Point cloud compression
Semi-supervised learning
Semi-supervised learning algorithm
Semisupervised learning
Three dimensional models
Three-dimensional displays
ISSN:2169-3536, 2169-3536
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:In autonomous vehicle, the traditional method of manually labeling laser point cloud data is not only expensive, but also complex in process. To overcome these challenges, a laser point cloud-image fusion technology based on semi-supervised learning algorithms is proposed to improve the efficiency and accuracy of 3D object detection. This study trains the model under partially supervised conditions to optimize detection accuracy and processing speed. The research results indicated that the semi-supervised learning algorithm had the lowest loss function value, which converged below 0.20 after 1000 iterations, far lower than other comparison algorithms. The root mean square error and mean absolute error of the semi-supervised learning algorithm were also the lowest, within 0.11. After testing on the KITTI, SUN-RGBD, and ScanNet v2 data sets, the proposed algorithm demonstrated a high accuracy of over 95%, demonstrating its generalization ability and robustness in different environments. The research provides an efficient and cost-effective 3D object detection solution for the field of autonomous driving, which has important practical application value.
Bibliografia:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2024.3461828