Systems and methods for controlling a vehicle using high precision and high recall detection
This disclosure provides systems and methods for controlling a vehicle based on a combination of high precision detection and high recall detection. The disclosed systems and methods can efficiently generate trajectories by reducing duplicate detection or duplicate calculation of objects or obstacles of common and known object types and objects or obstacles without class identification.
1 . A method of controlling an autonomous vehicle, comprising:
receiving data from a set of sensors, wherein the data represents objects or obstacles in an environment of the autonomous vehicle, and performing sensor fusion for the received data from the set of sensors; and
using a processor:
generating high precision detection data based on the received data;
identifying, from the high precision detection data, a first set of objects or obstacles that are classifiable by at least one known classifier;
tracking movement of one or more objects in the first set of objects or obstacles over time and maintaining identity of the tracked one or more objects in the first set of objects or obstacles;
generating high recall detection data based on the received data;
identifying from the high recall detection data a second set of objects or obstacles without using any classifier;
filtering out objects, from the second set of objects or obstacles, that correspond to the tracked one or more objects in the first set of objects or obstacles to obtain a filtered set of objects or obstacles;
determining a candidate trajectory for the autonomous vehicle to avoid at least the tracked one or more objects in the first set of objects or obstacles and the filtered set of objects or obstacles; and
controlling the autonomous vehicle according to the candidate trajectory.
2 . The method of claim 1 , comprising performing sensor fusion for the received data from a set of object detectors.
3 . The method of claim 1 , comprising generating the high precision detection data and the high recall detection data based on different subsets of data received from different sets of sensors.
4 . The method of claim 1 , wherein the step of filtering comprises filtering out a set of points corresponding to the tracked one or more objects in the first set of objects or obstacles from at least one point cloud of the second set of objects or obstacles.
5 . The method of claim 1 , comprising generating the high precision detection data from the data received from image and point cloud detectors.
6 . The method of claim 1 , comprising generating the high recall detection data from point cloud clustering by LiDAR, Stereo Depth Vision by RADAR, and/or Monocular Depth Vision using learned low-level features by RADAR.
7 . The method of claim 6 , herein the learned low-level features comprise distance.
8 . The method of claim 1 , wherein the step of tracking comprises tracking movement of the one or more objects in the first set of objects or obstacles over a period of time when the one or more objects are detectable by the set of sensors.
9 . The method of claim 8 , wherein the period is from about 100 ms to about 500 ms.
10 . The method of claim 1 , wherein the set of sensors comprise a two-dimensional object detector, a three-dimensional object detector, and/or an obstacle detector.
11 . The method of claim 1 , wherein the set of sensors comprise RADAR, LiDAR, camera, sonar, laser, or ultrasound.
12 . A system for controlling an autonomous vehicle, comprising:
a set of sensors, configured to receive data that represents objects or obstacles in an environment of the autonomous vehicle; and
a processor, configured to:
perform sensor fusion for the received data from the set of sensors;
generate high precision detection data based on the received data;
identify, from the high precision detection data, a first set of objects or obstacles that are classifiable by at least one known classifier;
track movement of one or more objects in the first set of objects or obstacles over time and maintain identity of the tracked one or more objects in the first set of objects or obstacles;
generate high recall detection data based on the received data;
identify from the high recall detection data a second set of objects or obstacles without using any classifier;
filtering out objects, from the second set of objects or obstacles, that correspond to the tracked one or more objects in the first set of objects or obstacles to obtain a filtered set of objects or obstacles;
determine a candidate trajectory for the autonomous vehicle to avoid at least the tracked one or more objects in the first set of objects or obstacles and the filtered set of objects or obstacles; and
control the autonomous vehicle according to the candidate trajectory.
13 . The system of claim 12 , wherein the processor is further configured to perform sensor fusion for the received data from a set of object detectors.
14 . The system of claim 12 , wherein the processor is configured to generate the high precision detection data and the high recall detection data based on different subsets of data received from different sets of sensors.
15 . The system of claim 12 , wherein the processor is configured to filter out a set of points corresponding to the tracked one or more objects in the first set of objects or obstacles from at least one point cloud of the second set of objects or obstacles.
16 . The system of claim 12 , wherein the processor is configured to generate the high precision detection data from the data received from image and point cloud detectors.
17 . The system of claim 12 , wherein the processor is configured to generate the high recall detection data from point cloud clustering by LiDAR, Stereo Depth Vision by RADAR, and/or Monocular Depth Vision using learned low-level features by RADAR.
18 . The system of claim 17 , wherein the learned low-level features comprise distance.
19 . The system of claim 12 , wherein the processor is configured to track movement of the one or more objects in the first set of objects or obstacles over a period of time when the one or more objects are detectable by the set of sensors.
20 . The system of claim 12 , wherein the set of sensors comprise a two-dimensional object detector, a three-dimensional object detector, and/or an obstacle detector.