Vehicle Lane Boundary Positioning Using Side Cameras

The patent replaces traditional computer-vision techniques with a neural network-based system that uses camera images to detect lane boundaries. The neural network is trained with human-annotated data to recognize various lane marking types, shadows, reflections, and road conditions, thereby improving detection accuracy while maintaining reasonable system complexity through automated learning rather than manual feature engineering.

The patent changes the approach from rule-based detection to learning-based detection by training a neural network on diverse training data representing different road conditions, lighting scenarios, and lane marking types. This parameter change in the detection methodology enables the system to adapt to variations in the environment, improving accuracy without requiring complex rule-based systems.

The patent performs preliminary action by training the neural network offline with extensive human-annotated training data before deployment. This pre-training allows the network to learn complex patterns and features, reducing the computational burden during real-time operation since the heavy lifting of pattern recognition has already been accomplished during the training phase.

The patent uses copying by creating a trained neural network model that can be deployed and reused multiple times. Once the network is trained with comprehensive data, the same model can be copied and deployed across different vehicles or systems, avoiding the need to retrain and reducing overall computational resource requirements.