Lidar Lane Marking Classification Using Global Scene Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to generate highly accurate maps of road elements, such as lane dividers and boundaries, from lidar scans due to difficulties in distinguishing between similar features and the computational complexity of combining lidar data with other sensor modalities.
Innovation Solution
A machine learning architecture using a convolutional neural network (CNN) processes pseudo-images from lidar scans to identify road markings, enhanced by a sub-network generating scene feature vectors for contextual information, enabling accurate distinction between similar features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If lidar data is combined with other sensor modalities to improve road element detection accuracy, then measurement precision improves, but device complexity increases
Solution Approach 1:
The patent extracts and processes only the necessary features from lidar data (intensity values, spatial coordinates, elevation) to create pseudo-images, rather than combining multiple sensor modalities. This selective extraction maintains measurement precision while reducing device complexity by eliminating the need for additional sensors.
Solution Approach 2:
The patent introduces pseudo-images as an intermediary representation that transforms raw lidar point cloud data into a format suitable for CNN processing. This intermediary structure enables accurate road element detection without requiring direct integration of multiple sensor types, thereby reducing system complexity.
2Ease of operation
If traditional image processing methods are used to identify road markings, then ease of operation is maintained, but measurement precision deteriorates due to inability to distinguish similar features
Solution Approach 1:
The patent replaces traditional mechanical/image processing methods with a neural network-based system that processes pseudo-images derived from lidar data. The CNN automatically learns to distinguish similar road markings through training, achieving high measurement precision while maintaining ease of operation through automated processing.
Solution Approach 2:
The patent transforms the input data representation by creating pseudo-images with specific parameters (intensity, spatial position, elevation) that enhance the distinguishability of similar road markings. This parameter transformation enables the neural network to achieve superior precision compared to traditional methods while keeping the processing pipeline straightforward.
3Measurement precision
If manual annotation of training data is performed to improve model accuracy, then measurement precision improves, but loss of time increases
Solution Approach 1:
The patent implements a self-service approach where the system uses its own detection outputs to automatically generate training data. The neural network processes lidar data and generates pseudo-images with automatic annotations based on detected road elements, eliminating the need for manual annotation while maintaining high measurement precision through iterative self-improvement.
Data Source
AI summary
Provided are methods, systems, and computer program products for generating an output map indicating a likelihood of individual elements of an image as corresponding to particular road elements, such as lane dividers, road dividers, and road boundaries. An example method may include applying a machine learning architecture to the image, which architecture includes a convolutional neural network and a sub-network capturing global context from feature maps generated by the convolutional neural network.