.png){kind=icon}
.png){kind=icon}
Return
3D-LiDAR point cloud data classification using deep learning for mobile robots
.png){kind=icon}
PRE
AI
DOI：10.1007/s10015-026-01115-8.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Navigation is an essential component of mobile robots. Typically, robots are equipped with visual sensors to identify objects in the vicinity for obstacle avoidance. However, visual sensors have several limitations like performance degradation in low illumination or depth perception. 3D LiDARs with many layers (32, 64, or 128), which generate dense point cloud data (PCD), have shown good performance in object detection to overcome such limitations. However, 3D LiDARs with fewer layers generate sparse data which are difficult for object detection. In this paper, sparse point cloud data are used for object identification using deep learning. Sparse PCD collected for different obstacles (person, multiple person, bicycle, and person walking with mobile) in left-right, and forward-backward configurations is augmented and passed into an neural network for identification. For robust testing, we have also included common immovable objects like chair. We achieved good results with sparse PCD for different obstacle configurations. The network can not only identify the type of obstacle but also the direction in which the obstacle is approaching. This information enables the robot path planner to make intelligent decisions for collision-free path planning. Data were collected using actual sensor and classification results are discussed.
Keywords:
Point cloud data
Robot navigation
Deep learning
Recognition
Journal
A
IF:
0.8
Papers:
74
Citations:
608