A Novel Virtual Navigation Route Generation Scheme for Augmented Reality Car Navigation System.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Automotive Technology

Background:

• Current augmented reality (AR) car navigation systems often rely on pre-rendered 3D content, leading to less authentic visual guidance.
• Complex road environments pose challenges for existing navigation systems in providing timely and accurate directions.

Purpose of the Study:

• To develop a novel virtual navigation route generation scheme for AR car navigation systems.
• To enhance the authenticity and accuracy of AR navigation effects.
• To improve driver guidance in complex traffic scenarios.

Main Methods:

• Utilized a generative adversarial network-long short-term memory (GAN-LSTM) framework.
• Employed an evolved fully convolutional network for accurate semantic segmentation of lane markings using inverse perspective mapping (IPM).
• Developed AR Navigation-Nets based on LSTM to predict virtual navigation routes and used a discriminator to ensure realism.

Main Results:

• Achieved "autonomous" generation of virtual navigation routes directly within captured images.
• Provided a more authentic and correct AR effect compared to superimposed 3D content.
• Enabled earlier and more accurate driver guidance in complex road traffic environments.

Conclusions:

• The proposed GAN-LSTM framework offers a significant advancement in AR car navigation.
• Autonomous virtual route generation enhances user experience and driving safety.
• The system effectively integrates semantic segmentation, path planning, and generative models for realistic navigation guidance.