LiDAR Pedestrian Segmentation Using a Logistic Cylinder Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately segmenting LiDAR points corresponding to pedestrians from nearby objects due to errors in label transfer between camera and LiDAR data, calibration issues, and degeneracies between 3D and 2D projections, leading to inefficient segmentation.
Innovation Solution
A system that projects point cloud data onto a 2D plane, determines optimized circle parameters to maximize probability distribution, and assigns pedestrian or non-pedestrian labels based on signed distance values, using brute force search or gradient ascent algorithms to efficiently classify data points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If graph-cuts are used to perform segmentation, then segmentation accuracy can be improved, but computational complexity increases and many tuning parameters are required
Solution Approach 1:
The patent extracts only the essential geometric features (cylinder center and radius) needed for pedestrian segmentation, removing the complex graph-cuts algorithm and its numerous tuning parameters. This extraction of key parameters simplifies the computational model while maintaining segmentation effectiveness.
Solution Approach 2:
The patent changes the parameter set from many graph-cuts tuning parameters to just two essential parameters (cylinder center and radius). This parameter reduction transforms the complex segmentation problem into a simpler geometric fitting problem that is computationally efficient while maintaining accuracy.
2Productivity
If simple partitioning is used to separate LiDAR points, then computational efficiency is improved, but segmentation accuracy deteriorates due to label transfer errors
Solution Approach 1:
The patent introduces a logistic cylinder pedestrian model as an intermediary between the raw LiDAR points and the final segmentation. This model acts as a mediator that incorporates prior knowledge about pedestrian geometry, allowing the system to correct label transfer errors while maintaining computational efficiency. The cylinder model serves as a bridge that refines the initial simple partitioning results.
3Measurement precision
If a logistic cylinder pedestrian model is used, then segmentation accuracy is improved, but the complexity of determining optimized parameters increases
Solution Approach 1:
The patent replaces complex iterative optimization mechanisms with a more straightforward parameter determination approach. Instead of using complex mechanical or iterative systems to find optimized parameters, the solution uses direct geometric fitting methods that are computationally simpler while achieving the same segmentation accuracy.
Data Source
AI summary
A system obtains, from an autonomous vehicle, point cloud data, projects the point cloud data onto a two-dimensional plane, and determines an optimized center parameter value of an optimized circle and an optimized radius parameter value of the optimized circle that, collectively, maximizes a probability distribution of a center parameter and a radius parameter across the point cloud data. The system determines whether one or more of the data points of the point cloud data are located within the optimized circle. If there are, the system assigns a pedestrian class value to the point label of the data point. If there are data points of the point cloud data that are located outside of the optimized circle, they system assigns a non-pedestrian class value to the point label of the data point.