| CPC G06V 20/58 (2022.01) [G06F 16/7837 (2019.01); G06N 20/00 (2019.01); G06T 7/292 (2017.01); G06V 20/46 (2022.01); G06V 20/48 (2022.01); G06V 20/56 (2022.01); G06N 3/08 (2013.01); G06T 2207/10016 (2013.01); G06T 2207/30252 (2013.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01); G07C 5/08 (2013.01); H04N 23/61 (2023.01)] | 14 Claims |

|
1. A computer-implemented method for associating labels in autonomous vehicle logs, the computer-implemented method comprising:
retrieving a logged track from one or more computing devices of an autonomous vehicle (AV) system, wherein the logged track comprises a sequence of logged frames;
retrieving a plurality of labeled tracks from a database, wherein each labeled track comprises a sequence of labeled frames;
for each logged frame in the logged track:
identifying a logged polygon in the respective logged frame; and for each labeled track of the plurality of labeled tracks:
retrieving a labeled frame that has a labeled timestamp within a predetermined time window around a logged timestamp of the logged polygon identified in the respective logged frame;
generating a similarity score based on a comparison of the logged polygon identified in the respective logged frame and a labeled polygon in the labeled frame;
identifying the respective logged frame and the labeled frame as a match-frame based on a comparison of the similarity score to a first user-defined threshold;
generating a candidate pair for each identified match-frame comprising the logged track and the labeled track associated with the identified match-frame;
removing one or more duplicate candidate pairs;
generating a match quality score for each of the candidate pairs;
generating a match quality score percentage for each of the candidate pairs by:
identifying the labeled track in the respective candidate pair;
identifying other candidate pairs that include the identified labeled track;
computing a sum of the match quality scores for the other candidate pairs that include the identified labeled track; and
dividing the match quality score for the respective candidate pair by the sum of the match quality scores for the other candidate pairs that include the identified labeled track; and
displaying the candidate pair based on a comparison of the match quality score to a second user-defined threshold.
|