System and method for training a machine-learning model to identify real-world elements

A method of training a neural network using a simulated environment (SE), the method comprising:

• creating at least one realistic element and at least one abstract element, wherein at least one realistic element is a realistic-portion, corresponding to at least one free-space in a non-simulated environment, and wherein at least one abstract-element is an abstract portion, corresponding to at least one occupied-space in a non-simulated environment and placing the elements in the SE;
• creating at least one synthetic image from the at least one SE;
• providing the at least one synthetic image to a neural network;
• training the neural network, to detect at least one free-space portion and at least one occupied-space portion in the SE, respectively corresponding to at least one free-space and at least one occupied-space in a non-simulated environment, based on the at least one synthetic image;
• and
• tagging at least one realistic-portion within the SE,
• and wherein training the neural network, to detect at least one free-space portion and at least one occupied-space portion in the non-simulated environment, comprises training the neural-network to generalize the at least one tagged realistic-portion to detect at least one corresponding free-space in a non-simulated environment.

A system for training a machine-learning model to identify real-world elements using a simulated environment (SE), the system comprising:

• a memory;
• and
• a processor configured to:
• receive at least one first set of appearance parameters corresponding to the appearance of at least one real-world element;
• generate one or more realistic elements, each corresponding to a variant of at least one real-world element, based on the at least one first set of appearance parameters;
• receive at least one second set of appearance parameters;
• generate one or more abstract elements according to the at least one second set of appearance parameters, wherein generating each of the one or more abstract elements further comprises applying a random value to at least one parameter of the at least one second set of appearance parameters to generate a variety of abstract elements;
• create at least one SE, and placing the one or more realistic elements and abstract elements within the SE;
• produce at least one synthetic image from the SE;
• provide the at least one synthetic image to a machine-learning model;
• train the machine-learning model to identify at least one real-world element from the at least one synthetic image that corresponds to at least one realistic element in the SE;
• and
• calculating at least one road delimiter through the free space portion of the SE, based on the location of at least one identified realistic element.