LiDAR Bounding Box Generation for Turning Vehicle Heading Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection methods using LiDAR sensors are inadequate for accurately determining the heading direction of nearby vehicles due to insufficient point data from the lateral side, leading to incorrect representation of the actual heading direction.
Innovation Solution
A method and system that determine whether an object is turning or straight-moving by analyzing LiDAR point cloud data, generate virtual points to complement sparse data, and create a bounding box based on these points to accurately represent the object's heading direction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional LiDAR detection method is used, then the detection process is simple, but the point data from lateral side of proximate vehicles is insufficient
Solution Approach 1:
The system performs preliminary classification of LiDAR points into front, rear, left, and right regions before bounding box generation. This preliminary organization of point data enables more comprehensive utilization of available points, particularly from lateral sides, to improve the quantity and distribution of points used in heading direction calculation without adding complex hardware
2Measurement precision
If conventional bounding box generation is used, then the processing is straightforward, but the detected heading cannot properly represent the actual heading direction of turning vehicles
Solution Approach 1:
The system applies different processing strategies to different regions of point data. Specifically, it classifies points into front, rear, left, and right regions and uses weighted combinations of bounding boxes from these regions. This local differentiation allows the system to account for turning vehicles by giving appropriate weight to lateral point regions, thereby improving heading direction accuracy without requiring complete redesign of the entire detection system
Solution Approach 2:
The bounding box generation process is segmented into multiple steps: generating front and rear bounding boxes, generating left and right bounding boxes, and then combining them with weighted coefficients. This segmentation allows the system to handle different regions independently and combine them to represent the overall vehicle heading, improving accuracy for turning vehicles while keeping each individual processing step manageable
3Quantity of substance
If more LiDAR sensors are added to capture lateral side points, then the point data coverage improves, but the system cost and complexity increase
Solution Approach 1:
The system processes point data in multiple dimensional regions (front, rear, left, right) and combines them with weighted coefficients to generate the final bounding box. This multi-dimensional approach allows comprehensive utilization of point cloud data from existing single LiDAR sensors, achieving improved lateral coverage without adding more sensors. The method transforms the problem from hardware expansion to software-based multi-region processing
Data Source
AI summary
The present disclosure relates to a method and a system for detecting an object. A method of detecting an object, according to an embodiment of the present disclosure, includes determining whether the object in a region of interest of a (host) vehicle is in a turning state or a straight moving state, determining a location of a reference point for the object based on the determination, extracting first outline points from LiDAR points for the object, which are included in a point cloud received from a LiDAR sensor, based on the reference point, generating at least one virtual point based on a length of a line segment connecting the first outline points, and generating a bounding box of the object corresponding to a heading direction of the object based on second outline points including the first outline points and the at least one virtual point.