| CPC G08G 1/166 (2013.01) [G01S 13/931 (2013.01); G08G 1/056 (2013.01)] | 19 Claims |
|
1. A method comprising:
maintaining a series of tracks including estimated measurements of at least some measurement detections output from an object detector model that processes sensor data to estimate portions of objects in an environment;
determining reported model confidence in the measurement detections based on estimates of uncertainties in the measurement detections for the objects in the environment;
updating estimated states of the tracks or generating new tracks based on a probability of existence (PoE) for the tracks and the reported model confidence in the measurement detections by,
associating, based on both the PoE for the tracks and the reported model confidence, each of the tracks with one or more of the measurement detections classified as high confidence measurement detections or the measurement detections classified as low confidence measurement detections such that at least some of the tracks are associated with the low confidence measurement detections, and
refining the estimated states of the tracks associated with low confidence measurement detections in subsequent update cycles using feedback from a fusion tracker incorporating ancillary information from other sensors to promote at least one of the tracks associated with the low confidence measurement detections to the high confidence measurement detections; and
electronically instructing a vehicle system to control a vehicle to avoid the objects in the environment based on the updated tracks or new tracks.
|