US 12,340,596 B2
Continuous context-based learning for lane prediction
Tom Tabak, Tel Aviv (IL)
Assigned to AUTOBRAINS TECHNOLOGIES LTD, Tel Aviv (IL)
Filed by AUTOBRAINS TECHNOLOGIES LTD, Tel Aviv-Jaffa (IL)
Filed on Oct. 11, 2022, as Appl. No. 18/045,824.
Claims priority of provisional application 63/262,389, filed on Oct. 11, 2021.
Prior Publication US 2023/0114215 A1, Apr. 13, 2023
Int. Cl. G06V 20/56 (2022.01); G06V 10/762 (2022.01); G06V 10/774 (2022.01)
CPC G06V 20/588 (2022.01) [G06V 10/762 (2022.01); G06V 10/774 (2022.01); G06V 2201/10 (2022.01)] 17 Claims
OG exemplary drawing
 
1. A method for lane boundary detection, the method comprises:
obtaining an image of an environment of a vehicle, the environment comprises at least one lane boundary portions; wherein the image comprises a first plurality of two dimensional image segments;
converting, by a first machine learning process, each two dimensional image segment of the first plurality of two dimensional image segments to a segment vector that represents the two dimensional image segment and comprises a sequence of pixels of the two dimensional image segment, to provide a first plurality of segment vectors;
finding an associated cluster for each of the first plurality of segment vectors to provide a second plurality of associated clusters;
searching for lane boundary relevant clusters of the second plurality of associated clusters, wherein the lane boundary relevant clusters comprise lane boundary relevant segment vectors that represent two dimensional image segments that are expected to include lane boundary pixels of lane boundary portions wherein the searching involves
locating the lane boundary portions within the two dimensional image segments, by:
determining, for each lane boundary relevant segment vector and by a second machine learning process, a location of the lane boundary portion within a two dimensional-image segment that is represented by the lane boundary relevant segment vector;
identifying a lane boundary irrelevant cluster of the second plurality of associated vectors; wherein the identified lane boundary irrelevant cluster comprises lane boundary irrelevant segment vectors that represent two dimensional segments that are not expected to include lane boundary pixels; and
skipping a search for a location of a lane boundary portion within any two dimensional image segment that is represented by a lane boundary irrelevant segment vector.