LiDAR Object Classification via Complex Coordinate FFT
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for autonomous driving struggle with accurately classifying objects in the vicinity of a vehicle equipped with a LiDAR sensor, which is crucial for effective autonomous driving.
Innovation Solution
A method and apparatus for classifying objects that involve transforming rectangular coordinates of points from a LiDAR point cloud into complex coordinates, performing Fast Fourier Transform (FFT) to extract a first feature, and obtaining an average and standard deviation as a second feature, based on which the type of object is classified.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object classification is performed using traditional LiDAR point cloud methods, then the system can detect objects, but the classification accuracy is insufficient for reliable autonomous driving
Solution Approach 1:
The patent transforms the coordinate system from rectangular to complex coordinates and applies Fast Fourier Transform to extract frequency domain features. This parameter transformation enables the system to capture subtle patterns in point cloud data that are not apparent in the spatial domain, thereby improving classification accuracy and reliability for autonomous driving applications.
Solution Approach 2:
The patent replaces traditional geometric-based classification methods with signal processing techniques (FFT and spectrum analysis). By treating the point cloud data as a signal and analyzing its frequency characteristics, the system achieves more robust object classification that is less sensitive to variations in lighting, occlusion, and sensor noise.
2Measurement precision
If FFT and spectrum analysis are applied to LiDAR point cloud data, then classification accuracy improves, but computational complexity increases
Solution Approach 1:
The patent performs coordinate transformation and FFT processing on the raw point cloud data immediately after data acquisition, before classification. By preprocessing the data in this manner, the system extracts meaningful frequency domain features that simplify subsequent classification tasks, making the overall process more efficient despite the added computational steps.
Solution Approach 2:
The patent extracts specific frequency domain features (spectrum characteristics) from the transformed data and uses only these extracted features for classification. This selective extraction of relevant features reduces the dimensionality of the problem and focuses computational resources on the most discriminative aspects of the data, balancing accuracy with complexity.
Data Source
AI summary
A method of classifying an object according to an embodiment includes extracting a first feature by transforming rectangular coordinates of points included in the box of the object, obtained from a point cloud acquired using a LiDAR sensor, into complex coordinates and performing Fast Fourier Transform (FFT) on the complex coordinates, obtaining an average and a standard deviation as a second feature, the average and the standard deviation being parameters of a Gaussian model for the points included in the box of the object, and classifying the type of object based on at least one of the first feature or the second feature.