Novel CE-CBCE feature extraction method for object classification using a low-density LiDAR point cloud.

Area of Science:

• Robotics and Automation
• Computer Vision
• Sensor Technology

Background:

• Low-end LiDAR sensors offer cost-effective depth sensing for lightweight devices.
• Limited computational resources on these devices restrict the use of complex algorithms for object recognition.
• Sparse data from low-end LiDAR hinders effective feature extraction.

Purpose of the Study:

• To develop a novel feature extraction method for sparse LiDAR point clouds.
• To enable accurate object classification on resource-constrained devices.
• To integrate feature extraction with a convolutional neural network (CNN) for efficient object recognition.

Main Methods:

• Proposed a Clustered Extraction and Centroid Based Clustered Extraction (CE-CBCE) method for feature extraction from sparse LiDAR data.
• Employed a Convolutional Neural Network (CNN) as the object classifier.
• Integrated CE-CBCE with CNN to leverage lightweight LiDAR input and low computing classification.

Main Results:

• The CE-CBCE method effectively extracts features from sparse LiDAR point clouds.
• The integrated CE-CBCE and CNN approach achieved reliable object detection.
• A classification accuracy of 97% was obtained using genuine LiDAR data.

Conclusions:

• The proposed CE-CBCE method is suitable for feature extraction from sparse LiDAR data.
• The integration of CE-CBCE and CNN provides an efficient solution for object recognition on lightweight devices.
• This approach maintains high detection accuracy under limited computational constraints.