AI-based autonomous vehicle path generation device and operating method using an around-view monitoring system

Autonomous Driving Path Generation Based on Around-View Images and Deep Neural Network Quadratic Function Estimation

This technology is a device and operating method for autonomous vehicles that extracts left and right lane coordinates from top-view images obtained via an Around-View Monitoring (AVM) system, estimates quadratic function coefficients for each lane using a trained deep neural network, and generates a path based on the median curve between the two.

Autonomous vehicles typically rely on front-facing cameras to detect lanes, making it difficult to maintain accurate paths in environments where lanes are short, broken, or curved, such as narrow alleys. AVM systems have historically been limited to providing visual assistance for parking or navigating tight spaces and have not been utilized for autonomous path generation.

This technology trains a deep neural network using lane coordinate sets and quadratic constant sets as training data to minimize loss, calculating curved paths based on lane contour center points detected while shifting the reference region of interest. Applicable to low-speed autonomous driving, automated parking, and alleyway navigation, it enables precise path generation that follows the center of the lane in narrow areas using only existing AVM cameras.

‍

‍

Key Features:

• A training data storage unit that stores sets of 2D coordinate values for lane paths and their corresponding quadratic function constant sets as machine learning training data
• A machine learning execution unit that generates an n x 2 input matrix from coordinate values, applies it to a deep neural network, and trains the network to minimize loss by comparing 3D output vectors with constant values
• A coordinate acquisition unit that obtains center point coordinates of left and right lane contours by shifting the reference region of interest in the AVM system's top-view image
• A path generation unit that creates the vehicle's movement path by calculating the median curve between the two quadratic function curves generated from the left and right output vectors

‍