Road Marking Detection Using 2D Images and Sparse 3D Point Clouds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting road traffic markings in car navigation and vehicle automatic driving systems are prone to inaccuracies due to sparse, occluded, or missing three-dimensional point clouds, and manual detection is inefficient and error-prone.
Innovation Solution
A data processing method that collects two-dimensional streetscape images and three-dimensional point clouds, performs region segmentation using inertial navigation data, and detects road traffic markings through binary processing and orthographic projection, reducing reliance on three-dimensional point clouds and improving accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If road traffic markings are detected directly from three-dimensional point cloud, then detection can be performed, but detection accuracy deteriorates when point cloud is sparse, occluded, or missing
Solution Approach 1:
The patent segments the detection process into two distinct pathways: one for processing two-dimensional streetscape images and another for processing three-dimensional point clouds. The image processing pathway extracts road traffic marking locations and colors, while the point cloud pathway extracts three-dimensional spatial information. These segmented results are then integrated to produce comprehensive detection output, thereby avoiding the accuracy degradation that occurs when relying solely on sparse or occluded point cloud data
Solution Approach 2:
The patent introduces an intermediary integration mechanism that combines detection results from two independent processing pathways (image-based and point cloud-based). This intermediary layer fuses the two-dimensional location and color information from images with the three-dimensional spatial information from point clouds, creating a more robust detection system that maintains accuracy even when one data source is degraded or incomplete
2Reliability
If manual detection method is used to extract road traffic markings from colorful point cloud, then detection can be performed, but processing efficiency deteriorates and mistakes increase
Solution Approach 1:
The patent replaces manual mechanical detection operations with automated computer vision algorithms. The system uses image processing techniques to automatically extract road traffic marking locations, colors, and three-dimensional coordinates from streetscape images and point cloud data, eliminating the need for manual extraction and thereby dramatically improving processing efficiency while maintaining or enhancing detection reliability
3Reliability
If ground reflectivity is weak or uneven, then detection may be performed, but detection accuracy deteriorates
Solution Approach 1:
The patent creates a universal detection system that processes multiple types of data (two-dimensional images and three-dimensional point clouds) through parallel pathways. This multi-functional approach allows the system to compensate for weaknesses in one data source using information from the other, ensuring reliable and accurate detection regardless of ground reflectivity conditions
Data Source
AI summary
At a computing system comprising one or more processors and memory, the computing system receives road data collected on a moving vehicle along a road, the road data comprising a two-dimensional streetscape image, a three-dimensional point cloud, and inertial navigation data, identifies, within the two-dimensional streetscape image, a ground region image corresponding to the road based on a spatial position relation of the two-dimensional streetscape image and the three-dimensional point cloud according to the inertial navigation data, and detects at least one target road traffic marking in the ground region image, determining three-dimensional coordinates of the at least one target road traffic marking based on the spatial position relation of the two-dimensional streetscape image and the three-dimensional point cloud.