Deep Learning Lane Line Detection for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional feature-based image segmentation models fail to detect whole lane lines and may predict lane lengths that are too short, leading to incomplete analysis and missed lane line detection in autonomous driving vehicles.
Innovation Solution
A deep learning image segmentation system that uses a convolutional neural network with weighted softmax cross-entropy loss to identify continuous lane lines from various lane markers, generating a trajectory for autonomous vehicle navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional feature-based image segmentation is used, then the system is simpler to implement, but it fails to detect whole lane lines and predicts too short lane lengths
Solution Approach 1:
The patent replaces traditional feature-based image segmentation methods with a deep learning-based convolutional neural network. This substitution enables the system to detect continuous lane lines effectively, resolving the contradiction between detection accuracy and system simplicity by using a more sophisticated approach that achieves superior performance.
Solution Approach 2:
The patent introduces parameter changes by modifying the loss function to include weighted softmax cross-entropy loss with specific weighting strategies. This allows the model to focus on detecting continuous lane lines while handling class imbalance, thereby improving detection accuracy without requiring excessive system complexity.
2Reliability
If traditional image segmentation is used, then processing is faster, but lane line detection is incomplete and unreliable
Solution Approach 1:
The patent replaces traditional image segmentation algorithms with a deep learning convolutional neural network. This substitution improves reliability by enabling continuous lane line detection, even though it requires more computational resources and processing time compared to traditional methods.
Solution Approach 2:
The patent implements preliminary action by pre-training the convolutional neural network with weighted softmax cross-entropy loss on labeled lane line data. This pre-training ensures the model learns effective features for continuous lane line detection, improving reliability while optimizing processing efficiency during deployment.
3Measurement precision
If the detection model predicts short lane segments, then local detection is more accurate, but overall lane line continuity is lost
Solution Approach 1:
The patent merges multiple detected lane line segments into continuous lane lines by using a convolutional neural network that processes the entire image context. The model combines local detection accuracy with global continuity information, resolving the contradiction between local precision and overall length by integrating features across the entire image.
Solution Approach 2:
The patent transitions from detecting discrete lane line segments to detecting continuous lane lines by changing the detection dimension. The convolutional neural network outputs probability maps that represent continuous lane lines, allowing the system to maintain local detection accuracy while achieving global continuity through a different representational dimension.
Data Source
AI summary
According to an embodiment, a system receives a captured image perceiving one or more lane markers surrounding the ADV from an image capturing device of the ADV. The system detects one or more continuous lane lines based on the one or more lane markers in the captured image by applying a machine learning model to the captured image, where the machine learning model includes a number of layers of nodes and the machine learning model includes a weighted softmax cross-entropy loss within at least one of the layers in training. The system generates a trajectory based on the one or more continuous (e.g., whole) lane lines to control the ADV autonomously according to the trajectory.