Vehicle LiDAR Object Classification via 3D-to-2D Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Inaccurate object information recognized by LiDAR sensors in autonomous driving systems can compromise safety and reliability, necessitating improved object detection and classification methods.
Innovation Solution
A vehicle LiDAR system that projects three-dimensional point clouds into two-dimensional images to extract feature information using a convolutional neural network (CNN), specifically employing a depth-wise separable convolution block for efficient object classification, including vehicles, pedestrians, and road boundaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional object detection methods are used with LiDAR sensors, then the system can detect objects, but the classification accuracy is insufficient leading to safety concerns
Solution Approach 1:
The patent transforms three-dimensional point cloud data from LiDAR into two-dimensional feature maps through projection operations. This dimensionality change enables the application of CNN algorithms traditionally designed for image data, significantly improving object classification accuracy while maintaining computational efficiency for autonomous driving safety applications.
2Measurement precision
If detailed feature extraction is performed to improve classification accuracy, then object recognition becomes more precise, but computational complexity increases
Solution Approach 1:
The patent extracts only the essential features needed for classification by projecting point cloud data into two-dimensional feature maps and using CNN to identify key patterns. This selective extraction of critical features achieves high classification accuracy while avoiding the computational burden of processing all raw point cloud data in full three-dimensional detail.
3Reliability
If comprehensive object information is processed to ensure accurate classification, then detection reliability improves, but processing time increases
Solution Approach 1:
The patent performs preliminary processing by projecting three-dimensional point cloud data into two-dimensional feature maps before classification. This pre-processing step organizes the data into a more computationally efficient format, allowing CNN to quickly process and classify objects with high reliability while reducing overall processing time for autonomous driving applications.
Data Source
AI summary
An embodiment object classification method for use with vehicle LiDAR systems includes projecting a three-dimensional point cloud acquired from an object by a LiDAR sensor into a two-dimensional image by extracting two-dimensional image-based feature information comprising shape information of the object and determining a type of the object by processing the two-dimensional image-based feature information based on a convolutional neural network.