Regression-Based Lane Line Detection With Spatial Detail Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional lane detection systems for autonomous vehicles suffer from loss of spatial information due to down-sampling in deep neural networks, leading to reduced precision and accuracy, and require significant computational resources and energy for real-time processing.
Innovation Solution
A regression-based line detection method that preserves spatial information by training machine learning models to predict distances and angles at the input resolution, allowing for precise recreation of lines at lower output resolutions, and uses bit encoding and clustering algorithms to reduce processing burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If down-sampling is performed during DNN processing to reduce computational load, then processing speed and energy efficiency are improved, but spatial information is lost leading to reduced detection precision and accuracy
Solution Approach 1:
The patent applies preliminary action by performing regression-based line detection on down-sampled images first, then using the detected line parameters (slope, intercept) to reconstruct precise lines on the original high-resolution image. This approach proactively handles the resolution issue by computing at lower resolution and transforming results back to high resolution, avoiding the need to process all pixels at full resolution while maintaining detection precision.
2Measurement precision
If separate output channels are used for each lane line class in DNN, then classification accuracy is improved, but computational resources and processing time are significantly increased
Solution Approach 1:
The patent changes the output parameters from multiple class-specific channels to a unified regression output that predicts line parameters (slope and intercept) for all lane lines. Instead of using separate output channels for each lane line class, the system transforms the classification problem into a regression problem where a single output layer predicts continuous line parameters, significantly reducing computational complexity while maintaining detection accuracy.
3Reliability
If post-processing steps are employed to reconstruct lanes from segmentation masks, then complete line information is recovered, but processing time is increased and accuracy is reduced due to working with lower resolution masks
Solution Approach 1:
The patent replaces the mechanical post-processing reconstruction approach with a direct regression-based line detection method. Instead of generating segmentation masks and then reconstructing lines through additional processing steps, the system directly regresses line parameters from the input image, eliminating the intermediate mask generation and reconstruction steps while working directly with high-resolution image data throughout the process.
Data Source
Figure 1A
Figure 1B
Figure 2A
AI summary
In various examples, systems and methods are disclosed that preserve rich spatial information from an input resolution of a machine learning model to regress on lines in an input image. The machine learning model may be trained to predict, in deployment, distances for each pixel of the input image at an input resolution to a line pixel determined to correspond to a line in the input image. The machine learning model may further be trained to predict angles and label classes of the line. An embedding algorithm may be used to train the machine learning model to predict clusters of line pixels that each correspond to a respective line in the input image. In deployment, the predictions of the machine learning model may be used as an aid for understanding the surrounding environment – e.g., for updating a world model – in a variety of autonomous machine applications.