Point Cloud Segmentation Using Depth Histograms for Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for classifying point group data from Lidar systems, such as Simple Euclidean Distance Based Segmentation, Difference of Normals, and Min-Cut Based Segmentation, face challenges in distinguishing adjacent objects and require high computational resources, making them unsuitable for real-time processing in autonomous driving applications.
Innovation Solution
A point group data processing device that acquires and processes depth information by setting target and enlargement areas, comparing histograms to accurately specify point groups corresponding to subjects, reducing computational load and enabling efficient classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If Simple Euclidean Distance Based Segmentation is used, then the classification process is simple, but the classification accuracy is insufficient and cannot distinguish adjacent objects
Solution Approach 1:
The patent changes the parameter basis from simple Euclidean distance to a composite parameter system including depth information, intensity information, and normal vector information. This allows the classification to consider multiple characteristics of point groups simultaneously, improving accuracy while maintaining reasonable computational complexity through efficient calculation methods for these parameters.
Solution Approach 2:
The patent uses temporary data structures (temporary point groups) that are created and discarded during the classification process. By forming temporary point groups based on depth and intensity information, performing classification, and then releasing them, the system achieves accurate classification without maintaining complex permanent data structures, balancing accuracy and computational efficiency.
2Measurement precision
If Difference of Normals (DoN) method is used, then the classification accuracy improves, but the calculation amount becomes enormous
Solution Approach 1:
Instead of calculating normal vectors for all points in the point cloud (excessive action), the patent calculates normal vectors only for points that are candidates for object boundaries or have significant geometric features (partial action). This selective approach maintains classification accuracy at boundaries while dramatically reducing the total calculation amount.
Solution Approach 2:
The patent applies different processing strategies to different regions of the point cloud. For example, it focuses computational resources on regions with high geometric variation or boundary characteristics where accurate normal vector calculation is most important, while using simpler methods for homogeneous regions. This local differentiation maintains accuracy where needed while reducing overall computation.
3Measurement precision
If Min-Cut Based Segmentation is used, then the classification accuracy improves, but the calculation amount becomes enormous and threshold selection becomes complex
Solution Approach 1:
The patent implements a two-stage segmentation approach: first forming temporary point groups based on depth and intensity information to create coarse segments, then performing refined classification within these segments using normal vector information. This hierarchical segmentation reduces the computational burden of the overall process while maintaining accurate boundary detection, avoiding the need for complex global threshold selection.
4Device complexity
If post-filtering processing with distance based clustering is used, then the processing is simplified, but adjacent objects cannot be distinguished
Solution Approach 1:
The patent performs preliminary classification based on depth and intensity information before final distance-based clustering. By pre-organizing points into temporary point groups with similar characteristics and identifying potential object boundaries in advance, the subsequent clustering process can distinguish adjacent objects more effectively without requiring complex filtering, thus maintaining simplicity while improving object distinction capability.
Data Source
AI summary
A point group data processing device includes: an image data acquisition unit configured to acquire a captured image; a point group data acquisition unit configured to acquire point group data indicating position information of a point group corresponding to a plurality of points included in the image; an area setting unit configured to set a target area which is an area surrounding a subject on the image and an enlargement area which is an area obtained by enlarging the target area; and a target point group specifying unit configured to specify a target point group corresponding to the subject based on depth information of a point group included in the target area and depth information of a point group included in the enlargement area, which are included in the point group data.