Lane Polyline Generation From Bird's-Eye Features for Direct Vehicle Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lane detection technologies for autonomous vehicles face challenges in accurately generating lane polylines, which are crucial for safe driving, and often require separate processing steps for control signals.
Innovation Solution
A neural network model is trained using bird's-eye-view features extracted from sensor data to generate lane polylines, incorporating seed probability loss, order loss, and embedding offset loss to correct pixel value imbalances and determine pixel orders, enabling end-to-end learning and direct control signal generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional lane detection methods are used, then lane detection can be performed, but additional separate processing steps are required to generate control signals
Solution Approach 1:
The patent combines lane detection and control signal generation into a single neural network model. The model directly outputs control signals (steering angle, acceleration, braking) from lane detection features, eliminating the need for separate processing steps and integrating multiple functions into one unified system.
Solution Approach 2:
The neural network model is designed to perform multiple functions simultaneously: it detects lanes, determines vehicle position relative to lanes, and generates control signals for steering, acceleration, and braking. This multi-functional approach reduces the number of separate components needed in the autonomous driving system.
2Measurement precision
If pixel value balancing is not applied, then training is simpler, but foreground and background pixel imbalances reduce detection accuracy
Solution Approach 1:
The patent applies parameter changes to the loss function by introducing weighted terms that account for foreground-background pixel imbalances. Different weights are assigned to foreground pixels (lane regions) and background pixels to ensure the model pays appropriate attention to each, improving detection accuracy without requiring complex data preprocessing.
3Measurement precision
If pixel order is not determined, then processing is faster, but accurate determination of vehicle position along the lane is compromised
Solution Approach 1:
The patent performs preliminary ordering of pixels along the lane direction during the training phase. By pre-establishing the spatial order relationships between pixels using the determined pixel order, the model can quickly estimate vehicle position without performing complex calculations during real-time operation, thus reducing processing time while maintaining accuracy.
Data Source
AI summary
The present disclosure relates to a method and device for generating a lane polyline by using a neural network model.The method according to an embodiment of the present disclosure may include generating a bird's-eye-view feature based on a base image obtained from at least one sensor mounted on a vehicle, and training a neural network model by using the bird's-eye-view feature as input data for the neural network model and using a lane polyline for a certain road as output data. In the present disclosure, a lane polyline obtained from the above-described neural network model may be used for controlling the vehicle without performing a separate process on the lane polyline.