BAYESIAN NETWORK TO TRACK OBJECTS USING SCAN POINTS USING MULTIPLE LiDAR SENSORS

• US 20130246020A1
• Filed: 03/07/2013
• Published: 09/19/2013
• Est. Priority Date: 03/15/2012
• Status: Active Grant

• Current Assignee: GM Global Technology Operations LLC (General Motors Company)
• Original Assignee: GM Global Technology Operations LLC (General Motors Company)
• Inventors: ZENG, SHUQING
• Granted Patent: US 9,129,211 B2
• Time in Patent Office: Days
• US Class Current: 703/2
• CPC Class Codes: G01S 17/66 Tracking systems using electromagnetic waves other than radio waves
• CPC Class Codes: G01S 17/86 Combinations of lidar systems with systems other than lidar, radar or sonar, e.g. with direction finders
• CPC Class Codes: G01S 17/87 Combinations of systems using electromagnetic waves other than radio waves
• CPC Class Codes: G01S 17/931 of land vehicles
• CPC Class Codes: G01S 7/4808 Evaluating distance, position or velocity data
• CPC Class Codes: G06F 30/20 Design optimisation, verification or simulation optimisation, verification or simulation of circuit designs G06F30/30
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A system and method for fusing the outputs from multiple LiDAR sensors on a vehicle. The method includes providing object files for objects detected by the sensors at a previous sample time, where the object files identify the position, orientation and velocity of the detected objects. The method also includes receiving a plurality of scan returns from objects detected in the field-of-view of the sensors at a current sample time and constructing a point cloud from the scan returns. The method then segments the scan points in the point cloud into predicted clusters, where each cluster initially identifies an object detected by the sensors. The method matches the predicted clusters with predicted object models generated from objects being tracked during the previous sample time. The method creates new object models, deletes dying object models and updates the object files based on the object models for the current sample time.

Claims

1. A method for fusing outputs from multiple LiDAR sensors on a vehicle, said method comprising:

providing object files for objects detected by the LiDAR sensors at a previous sample time, said object files having object models that identify a position, orientation and velocity of the objects detected by the sensors;

tracking the object models of the objects detected by the sensors;

projecting tracked object models in the object files from the previous scan time to provide predicted object models;

receiving a plurality of scan returns from objects detected in a field-of-view of the sensors at a current sample time;

constructing a scan point cloud from the scan returns;

segmenting the scan points in the point cloud into predicted scan clusters where each scan cluster initially identifies an object detected by the sensors;

matching the predicted scan clusters with the predicted object models;

merging predicted object models that have been identified as separate scan clusters in the previous sample time but are now identified as a single scan cluster in the current sample time;

splitting predicted object models that have been identified as a single scan cluster in the previous sample time but are now identified as separate scan clusters in the current sample time;

creating new object models for detected objects in the current sample time that were not present in the previous sample time;

deleting object models that are no longer present in the predicted scan clusters;

providing object model updates based on the merged object models, split object models, new object models and deleted object models; and

updating the object files for the current sample time with the object model updates.

2. The method according to claim 1 wherein segmenting the scan points in the point cloud includes separating the clusters of scan points in the point cloud so that the clusters identify a separate object that is being tracked.

3. The method according to claim 1 wherein matching the predicted scan clusters with the predicted object models includes generating a bipartite graph that matches the scan clusters to the predicted object models.

4. The method according to claim 3 wherein generating a bipartite graph includes positioning the scan points at vertices of the graph.

5. The method according to claim 3 wherein merging predicted object models and splitting predicted object models includes converting the bipartite graph to an induced bipartite graph showing the object models that are being merged or split.

6. The method according to claim 5 wherein converting the bipartite graph to an induced bipartite graph includes computing weights and a cardinality of edges for the induced bipartite graph, where each weight identifies a positional change of one of the scan points for the at least one sensor to a location of a scan point for another one of the sensors and the edges define matching between object model points and segmented scan points.

7. The method according to claim 6 wherein converting the bipartite graph to an induced bipartite graph includes comparing the cardinality of the edges to a threshold and highlighting the edge if the cardinality is greater than the threshold, and wherein an edge is removed if it not highlighted.

8. The method according to claim 7 wherein deleting object models that are no longer present in the predicted scan clusters includes deleting an object model that was connected to an edge that is removed.

9. The method according to claim 1 wherein providing object model updates includes identifying the target data, a transformation parameter, the object models and the scan map clusters in a Bayesian network from the previous time frame to the current time frame.

10. The method according to claim 1 wherein updating the object files includes revising a transformation of the sensors that identifies an orientation angle and position of the sensor.

11. A method for fusing outputs from multiple LiDAR sensors on a vehicle, said method comprising:

providing object files for objects detected by the LiDAR sensors at a previous sample time, said object files having object models that identify a position, orientation and velocity of the objects detected by the sensors;

tracking the object models of the objects detected by the sensors;

projecting tracked object models in the object files from the previous scan time to provide predicted object models;

receiving a plurality of scan returns from objects detected in a field-of-view of the sensors at a current sample time;

constructing a scan point cloud from the scan returns;

segmenting the scan points in the point cloud into predicted scan clusters where each scan cluster initially identifies an object detected by the sensors;

matching the predicted scan clusters with the predicted object models;

merging predicted object models that have been identified as separate scan clusters in the previous sample time but are now identified as a single scan cluster in the current sample time; and

splitting predicted object models that have been identified as a single scan cluster in the previous sample time but are now identified as separate scan clusters in the current sample time.

12. The method according to claim 11 wherein segmenting the scan points in the point cloud includes separating the clusters of scan points in the point cloud so that the clusters identify a separate object that is being tracked.

13. The method according to claim 11 wherein matching the predicted scan clusters with the predicted object models includes generating a bipartite graph that matches the scan clusters to the predicted object models.

14. The method according to claim 13 wherein generating a bipartite graph includes positioning the scan points at vertices of the graph.

15. The method according to claim 13 wherein merging predicted object models and splitting predicted object models includes converting the bipartite graph to an induced bipartite graph showing the object models that are being merged and split.

16. The method according to claim 15 wherein converting the bipartite graph to an induced bipartite graph includes computing weights and a cardinality of edges for the induced bipartite graph, where each weight identifies a positional change of one of the scan points for the at least one sensor to a location of a scan point for another one of the sensors and the edges define matching between object model points and segmented scan points.

17. The method according to claim 16 wherein converting the bipartite graph to an induced bipartite graph includes comparing the cardinality of the edges to a threshold and highlighting the edge if the cardinality is greater than the threshold, and wherein an edge is removed if it not highlighted.

18. A system for fusing outputs from multiple LiDAR sensors on a vehicle, said system comprising:

means for providing object files for objects detected by the LiDAR sensors at a previous sample time, said object files having object models that identify a position, orientation and velocity of the objects detected by the sensors;

means for tracking the object models of the objects detected by the sensors;

means for projecting tracked object models in the object files from the previous scan time to provide predicted object models;

means for receiving a plurality of scan returns from objects detected in a field-of-view of the sensors at a current sample time;

means for constructing a scan point cloud from the scan returns;

means for segmenting the scan points in the point cloud into predicted scan clusters where each scan cluster initially identifies an object detected by the sensors;

means for matching the predicted scan clusters with the predicted object models;

means for merging predicted object models that have been identified as separate scan clusters in the previous sample time but are now identified as a single scan cluster in the current sample time;

means for splitting predicted object models that have been identified as a single scan cluster in the previous sample time but are now identified as separate scan clusters in the current sample time;

means for creating new object models for detected objects in the current sample time that were not present in the previous sample time;

means for deleting object models that are no longer present in the predicted scan clusters;

means for providing object model updates based on the merged object models, split object models, new object models and deleted object models; and

means for updating the object files for the current sample time with the object model updates.

19. The system according to claim 18 wherein the means for matching the predicted scan clusters with the predicted object models generates a bipartite graph that matches the scan clusters to the predicted object models.

20. The system according to claim 19 wherein the means for merging predicted object models and splitting predicted object models converts the bipartite graph to an induced bipartite graph showing the object models that are being merged and split.