Related Experiment Video
Updated: Oct 13, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Lane and Road Marker Semantic Video Segmentation Using Mask Cropping and Optical Flow Estimation
1School of Automation and Electronic Engineering, Qingdao University of Science and Technology, Qingdao 266061, China.
Sensors (Basel, Switzerland)
|November 13, 2021
Summary
This study introduces a deep learning algorithm for stable lane and road marker segmentation in autonomous driving. It improves accuracy and speed by using adjacent frame continuity and optical flow, enhancing real-time performance.
Area of Science:
- Computer Vision
- Deep Learning
- Autonomous Driving Systems
Background:
- Lane and road marker segmentation is vital for autonomous driving safety.
- Existing single-frame methods lack temporal stability, while multi-frame approaches suffer from error accumulation and slow speeds.
Purpose of the Study:
- To develop a deep learning algorithm for robust and efficient lane and road marker segmentation.
- To enhance temporal consistency and accuracy in video-based segmentation.
Main Methods:
- Proposed an end-to-end trainable network processing image sequences.
- Incorporated temporal consistency by expanding previous frame segmentation and using reversed optical flow as additional input.
- Utilized adjacent frame continuity and optical flow for improved target area attention.
Main Results:
- Achieved a 2.5x acceleration in segmentation speed for video lanes and road markers.
- Increased segmentation accuracy by 1.4%.
- Maintained high temporal consistency, with a maximum reduction of only 2.2%.
Conclusions:
- The proposed deep learning algorithm significantly improves the speed and accuracy of lane and road marker segmentation.
- The method effectively leverages temporal information from adjacent frames for enhanced performance in autonomous driving applications.
