METHOD AND SYSTEM FOR TRAINING MACHINE LEARNING ALGORITHM FOR DETECTING OBJECTS AT A DISTANCE Russian patent published in 2022 - IPC G06K9/62 G06N20/00 

FIELD: machine learning.

SUBSTANCE: invention relates to methods and a server for training a machine learning algorithm (hereinafter – MLA) to detect objects in sensor data. The method for training the first machine learning algorithm (MLA) to detect objects in sensor data received by the second sensor located at the second distance from objects is described, where the first MLA algorithm is trained to recognize objects in sensor data received by the first sensor located at the first distance from objects, and the second distance is greater than the first distance. The method is performed on the server and includes: reception by the server of the first sensor data received at the first distance by the first sensor installed on the first vehicle that passed along the predetermined road; reception by the server of the second sensor data received at the second distance by the second sensor installed on the second vehicle that passed behind the first vehicle along the predetermined road; reception by the server of a three-dimensional map of the predetermined road, including a set of objects; combination by the server of the first sensor data with the second sensor data, based, at least partially, on the set of objects on the three-dimensional map, to obtain combined first sensor data and combined second sensor data so that the given area in combined first sensor data corresponds to the given area in combined second sensor data; determination by the first MLA algorithm of parts of combined first sensor data corresponding to objects, including the determination of classes of these objects; establishment by the server of the compliance of objects and classes of these objects with parts of combined second sensor data corresponding to parts of combined first sensor data; training by the server of the first MLA algorithm to detect objects and classes of these objects in sensor data received at the second distance from objects, based on parts of combined second sensor data and classes of objects, for which compliance with parts of combined second sensor data is established.

EFFECT: possibility of training MLA algorithm to recognize objects with a less accurate representation of these objects in sensor data.

25 cl, 8 dwg