3D Lane Boundary Modeling Across Traversals and Occlusions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems face challenges in transferring lane boundary information across different traversals and modalities due to occlusions and the difficulty in accurately detecting lane boundaries using machine learning models.
Innovation Solution
A method involving the use of a three-dimensional LiDAR point cloud and a machine learning model to generate a lane boundary model, where LiDAR points are paired with images to construct a spline of best fit for lane boundaries, and a further model is used to remove occlusions, enabling accurate lane boundary detection across different traversals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If a machine learning model is used to detect lane boundaries in images, then lane boundary detection can be automated, but the detected lane boundaries are difficult to transfer to other traversals and modalities
Solution Approach 1:
The patent introduces a LiDAR point cloud as an intermediary representation between the image and the lane boundary model. The machine learning model detects lane boundaries in the image, these detections are then transferred to the LiDAR point cloud to generate a modality-agnostic lane boundary model. This intermediary representation enables automated detection while improving transferability to other traversals and modalities.
Solution Approach 2:
The patent transitions from two-dimensional image data to three-dimensional LiDAR point cloud data. By projecting the detected lane boundaries onto the 3D point cloud and generating a lane boundary model in 3D space, the system creates a representation that is independent of the original image modality and can be transferred to other traversals more effectively.
2Device complexity
If lane boundaries are detected from images, then the detection process is simple, but occlusions make it difficult to accurately detect lane boundaries
Solution Approach 1:
The patent merges information from multiple sources: the machine learning model's lane boundary detection in the image, the LiDAR point cloud data, and the spline fitting process. By combining these multiple information sources, the system overcomes the limitations of image-only detection and achieves more accurate lane boundary detection even in the presence of occlusions.
Solution Approach 2:
The patent creates a copy of the lane boundary information in a different modality. The lane boundaries detected in the image are projected onto the LiDAR point cloud, creating a modality-agnostic representation that is less susceptible to occlusions and can be used for more accurate detection.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method allows for easy transfer of lane boundary models to other images and modalities, improving the accuracy and reliability of lane boundary detection in autonomous vehicles by addressing occlusions.
Implementation Method 1
obtaining a three-dimensional LiDAR point cloud of a route
Data Source
AI summary
A computer-implemented method of generating a lane boundary model of a route traversed by an autonomous vehicle According to the present invention, there is provided a computer-implemented method of generating a lane boundary model of a route traversed by an autonomous vehicle. The computer-implemented method comprises: obtaining a three-dimensional LiDAR point cloud of a route and an image of the route traversed by the autonomous vehicle; detecting, using a machine learning model, a lane boundary in the image of the route; and generating a lane boundary model based on a plurality of points of the three-dimensional LiDAR point cloud of the route that correspond positionally with the lane boundary detected in the time.