US 11,703,562 B2
Semantic segmentation of radar data
Ankit Laddha, Pittsburgh, PA (US); Carlos Vallespi-Gonzalez, Wexford, PA (US); Duncan Blake Barber, Pittsburgh, PA (US); Jacob White, Glenshaw, PA (US); and Anurag Kumar, Pittsburgh, PA (US)
Assigned to UATC, LLC, Mountain View, CA (US)
Filed by UATC, LLC, San Francisco, CA (US)
Filed on Sep. 19, 2019, as Appl. No. 16/575,855.
Claims priority of provisional application 62/870,998, filed on Jul. 5, 2019.
Prior Publication US 2021/0003665 A1, Jan. 7, 2021
Int. Cl. G01S 13/86 (2006.01); G01S 7/28 (2006.01); G01S 13/931 (2020.01); G01S 13/90 (2006.01)
CPC G01S 7/2813 (2013.01) [G01S 13/865 (2013.01); G01S 13/9011 (2013.01); G01S 13/9017 (2013.01); G01S 13/931 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method of sensor output segmentation, the computer-implemented method comprising:
accessing sensor data comprising a plurality of sensor data returns representative of an environment detected by at least one sensor across a field of view of the at least one sensor;
associating the plurality of sensor data returns with a plurality of bins corresponding to the field of view of the at least one sensor, wherein the plurality of bins respectively correspond to a different angular portion associated with a number of degrees of the field of view of the at least one sensor;
for a bin of the plurality of bins that contains multiple sensor data returns respectively indicating objects at multiple different distances, selecting a sensor data return of the multiple sensor data returns in the bin, the selected sensor data return being characterized by a minimum distance of the multiple different distances;
generating a plurality of channels for the plurality of bins, the plurality of channels respectively comprising data indicative of a range and an azimuth associated with at least one sensor data return associated with a respective bin of the plurality of bins, and wherein the plurality of channels further comprises data indicative of a signal to noise ratio associated with the sensor data of respective bins of the plurality of bins; and
generating a semantic segment of at least a portion of the sensor data representative of the environment by inputting the plurality of channels into a machine-learned segmentation model trained to segment at least a portion of the plurality of sensor data returns based at least in part on input comprising the plurality of channels, wherein the machine-learned segmentation model generates at least one output comprising the semantic segment.