Hybrid Lane Estimation Using Deep Learning and Computer Vision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous and semi-autonomous driving systems face challenges in accurately estimating lane boundaries, with deep learning-based methods being slow and having poor vertex-level accuracy, while computer vision-based methods are fast but limited to short and narrow lane detection.
Innovation Solution
A hybrid lane estimation method that combines deep learning and computer vision by assigning camera frames to both detectors based on availability, leveraging the strengths of each to improve lane boundary detection and model generation, and updating previously identified lane models for enhanced accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep learning-based lane detection is used, then long-range and occluded lane detection capability is improved, but processing speed deteriorates
Solution Approach 1:
The patent segments the lane detection task by dividing camera frames into different subsets processed by DL and CV detectors respectively. The system assigns certain frames to DL detector for high-reliability detection and other frames to CV detector for fast processing, thereby resolving the contradiction between reliability and speed.
Solution Approach 2:
The patent merges deep learning-based lane detector and computer vision-based lane detector into a hybrid system. Both detectors work together with their respective strengths combined - DL provides long-range and occluded lane detection while CV provides fast processing, achieving both high reliability and high productivity.
2Productivity
If computer vision-based lane detection is used, then processing speed is improved, but detection range is limited to short and narrow lanes
Solution Approach 1:
The patent segments the detection task by assigning different frame subsets to CV and DL detectors. The CV detector processes frames where fast processing is sufficient, while DL detector handles frames requiring extended detection range, thus achieving both high speed and comprehensive coverage.
Solution Approach 2:
The patent combines CV-based detector with DL-based detector to merge their detection capabilities. The CV detector provides fast processing for visible lanes while the DL detector extends detection to long-range and occluded lanes, together achieving both high speed and wide detection range.
3Area of stationary object
If deep learning-based lane detection is used, then long-range detection is improved, but vertex-level accuracy deteriorates
Solution Approach 1:
The patent segments the detection task by assigning frames requiring long-range detection to the DL detector and frames requiring high vertex-level accuracy to the CV detector. This segmentation allows each detector to operate in its optimal performance zone.
Solution Approach 2:
The patent merges the results from DL detector and CV detector through a lane tracker. The DL detector provides long-range detection capability while the CV detector provides accurate vertex-level detection, and their combined output achieves both extended range and high precision.
Data Source
AI summary
Disclosed are techniques for lane estimation. In aspects, a method includes receiving a plurality of camera frames captured by a camera sensor of a vehicle, assigning a first subset of the plurality of camera frames to a deep learning (DL) detector and a second subset of the plurality of camera frames to a computer vision (CV) detector based on availability of the DL and CV detectors, identifying a first set of lane boundary lines in a first camera frame processed by the DL detector, identifying a second set of lane boundary lines in a second camera frame processed by the CV detector, generating first and second sets of lane models based on the first and second sets of lane boundary lines, and updating a set of previously identified lane models based on the first set of lane models and/or the second set of lane models.