EP-4181088-B1 - CLUSTERING TRACK PAIRS FOR MULTI-SENSOR TRACK ASSOCIATION

EP4181088B1EP 4181088 B1EP4181088 B1EP 4181088B1EP-4181088-B1

Inventors

IMRAN, SYED ASIF

Dates

Publication Date: 20260513
Application Date: 20220926

Claims (13)

A system comprising: at least one processor configured to: determine (502) candidate track pairs, each candidate track pair representing a connection between a first track obtained by a first sensor and a second track obtained by a second sensor, the connection based on an association between the first track and the second track to represent a same object; determine, from a perspective of the tracks obtained by the first sensor compared to a perspective of the tracks obtained by the second sensor, a strength of the connection associated with each of the candidate track pairs; responsive to identifying, based on the strength of the connection associated with each of the candidate track pairs, a connection with a weakest strength associated with the candidate track pairs, prune (504) a weakest candidate track pair, the weakest candidate track pair having the connection with the weakest strength; responsive to pruning the weakest candidate track pair, cluster (506) the candidate track pairs into at least a first cluster of the candidate track pairs and a second cluster of the candidate track pairs; determine (508), based on track-association techniques, a set of object track pairs from the at least first cluster of the candidate track pairs and the second cluster of the candidate track pairs, a first track and a second track of each respective object track pair in the set of object track pairs representing the same object; and output (510) the set of object track pairs to an automotive system of a vehicle to control operations of the vehicle.
The system of claim 1, wherein the at least one processor is configured to identify the connection with the weakest strength associated with the candidate track pairs by at least: determining, a feasibility of each candidate track pair based on a likelihood that the first track and the second track of the respective candidate track pair represents the same object; determining, based on the feasibility of the connection represented by each feasible candidate track pair, an evidence of the connection associated with each feasible candidate track pair; responsive to determining the evidence of the connection associated with each feasible candidate track pair: comparing, to one another, the evidence of the connections associated with each candidate track pair having the same first track; and comparing, to one another, the evidence of the connections associated with each candidate track pair having the same second track; and responsive to comparing the evidence of the connections associated with each candidate track pair having the same first track and comparing the evidence of the connections associated with each candidate track pair having the same second track, identifying the weakest candidate track pair.
The system of claim 2, wherein the at least one processor is configured to compare the evidence of the connections associated with each candidate track pair having the same first track and compare the evidence of the connections associated with each candidate track pair having the same second track by at least: normalizing the evidence of the connection of each first track with the second tracks that are feasible with the respective first track, and normalizing the evidence of the connection of each second track with the first tracks that are feasible with the respective second track.
The system of claim 3, wherein the at least one processor is further configured to: combine, for each respective candidate track pair, the normalized evidence based on the connection of each first track with the second tracks that are feasible with the respective first track and the normalized evidence based on of the connection of each second track with the first tracks that are feasible with the respective second track.
The system of claim 4, wherein the connection of the weakest candidate track pair is identified based on the combined normalized evidence associated with each candidate track pair.
The system of claim 5, wherein the at least one processor is further configured to prune the weakest candidate track pair by at least: pruning any candidate track pair with a connection having a combined normalized evidence below a threshold.
The system of claims 5 or 6, wherein the at least one processor is further configured to: maintain a feasibility matrix including feasibility data for each candidate track pair.
The system of claim 7, wherein the at least one processor is further configured to: maintain a cost matrix including cost data for each candidate track pair that is feasible, the cost data based on comparing data obtained by the first sensor to data obtained by the second sensor to determine a measure of association between the first track and the second track of the candidate track pair.
The system of claim 8, wherein the at least one processor is further configured to: maintain an evidence matrix including evidence data determined by inverting the cost data.
The system of claim 9, wherein the at least one processor is further configured to: maintain a connection weight matrix including the combined normalized evidence associated with each candidate pair.
The system of any one of the preceding claims, wherein the track-association techniques comprise at least one of a joint probabilistic association filter (JPDA) or multiple hypothesis tracking (MHT).
A computer-readable storage medium comprising instructions that, when executed, configures a processor to: determine (502) candidate track pairs, each candidate track pair representing a connection between a first track obtained by a first sensor and a second track obtained by a second sensor, the connection based on an association between the first track and the second track to represent a same object; determine, from a perspective of the tracks obtained by the first sensor compared to a perspective of the tracks obtained by the second sensor, a strength of the connection associated with each of the candidate track pairs; responsive to identifying, based on the strength of the connection associated with each of the candidate track pairs, a connection with a weakest strength associated with the candidate track pairs, prune (504) a weakest candidate track pair, the weakest candidate track pair having the connection with the weakest strength; responsive to pruning the weakest candidate track pair, cluster (506) the candidate track pairs into at least a first cluster of the candidate track pairs and a second cluster of the candidate track pairs; deteremine (508), based on track-association techniques, a set of object track pairs from the at least first cluster of the candidate track pairs and the second cluster of the candidate track pairs, a first track and a second track of each respective object track pair in the set of object track pairs representing the same object; and output (510) the set of object track pairs to an automotive system of a vehicle to control operations of the vehicle.
A computer-implemented method comprising: determining (502) candidate track pairs, each candidate track pair representing a connection between a first track obtained by a first sensor and a second track obtained by a second sensor, the connection based on an association between the first track and the second track to represent a same object; determining, from a perspective of the tracks obtained by the first sensor compared to a perspective of the tracks obtained by the second sensor, a strength of the connection associated with each of the candidate track pairs; responsive to identifying, based on the strength of the connection associated with each of the candidate track pairs, a connection with a weakest strength associated with the candidate track pairs, pruning (504) a weakest candidate track pair, the weakest candidate track pair having the connection with the weakest strength; responsive to pruning the weakest candidate track pair, clustering (506) the candidate track pairs into at least a first cluster of the candidate track pairs and a second cluster of the candidate track pairs; determining (508), based on track-association techniques, a set of object track pairs from the at least first cluster of the candidate track pairs and the second cluster of the candidate track pairs, a first track and a second track of each respective object track pair in the set of object track pairs representing the same object; and outputting (510) the set of object track pairs to an automotive system of a vehicle to control operations of the vehicle.

Description

BACKGROUND Automotive vehicles are becoming more sophisticated with the addition of sensors used to track objects near the vehicle. These objects may include other vehicles, pedestrians, animals, and inanimate objects, such as trees and street signs. The sensors (e.g., optical cameras, radar, Light Detection and Ranging (LiDAR)) collect low-level data (e.g., low-level tracks) that can be used to infer position, velocity, trajectory, class, and other parameters of the objects. A cluttered environment may result in many tracks from the various sensors. As the tracks are generated, steps may be taken to pair and associate the tracks originating from the different sensors to a common object. With many different possible associations or pairings of tracks, correctly matching the different tracks to a common object can be computationally complex. The computational complexity increases exponentially as the quantity of low-level tracks increases. Improving the processing speed (e.g., reducing the computational complexity) required for track association may increase the safety and reliability of an automobile, in particular, for autonomous or semi-autonomous control. The Article "Efficient Track-to-Track Assignment Using Cluster Analysis" by A. L. Bereson and R. N. Lobbia (In: 2006 9th International Conference on Information Fusion (pp. 1-8). IEEE, ISBN: 978-1-4244-0953-2) describes a Track-To-Track (T2T) data fusion system which uses an agglomerative hierarchical clustering algorithm. A chi-square gate test is used to determine which pairs of tracks may be assigned to one another. Assignment scores for pairs that do not pass this gating are pruned prior to clustering. The article "Track-to-Track Association for Intelligent Vehicles by Preserving Local Track Geometry" by Zou Ke et al (Sensors, vol. 20, no. 5, 1 March 2020, page 1412, ISSN: 1424-8220, DOI: 10.3390/s20051412) proposes a probabilistic method, called the local track geometry preservation (LTGP) algorithm, which takes advantage of the geometry of tracks. Assuming that the local tracks of one sensor are represented by Gaussian mixture model (GMM) centroids, the corresponding local tracks of the other sensor are fitted to those of the first sensor. A geometrical descriptor connectivity matrix is constructed to exploit the relative geometry of these tracks. The track association problem is formulated as a maximum likelihood estimation problem with a local track geometry constraint, and an expectation-maximization (EM) algorithm is developed to find the solution. SUMMARY This document describes systems and techniques for clustering track pairs for multi-sensor track association. Many track-association algorithms use pattern-matching processes that can be computationally complex. Clustering tracks derived from different sensors present on a vehicle may reduce the computational complexity by reducing the pattern-matching problem into groups of subproblems. The weakest connection between two sets of tracks is identified based on both the perspective from each track derived from a first sensor and the perspective of each track derived from a second sensor. By identifying and pruning the weakest connections between two sets of tracks, a large cluster of tracks may be split into smaller clusters. The smaller clusters may require fewer computations by limiting the quantity of candidate track pairs to be evaluated. Fewer computations result in processing the sensor information more efficiently that, in turn, may increase the safety and reliability of an automobile. Aspects described below include clustering track pairs for multi-sensor track association. In one example, a system includes at least one processor that can determine candidate track pairs, each candidate track pair representing a connection between a first track obtained by a first sensor and a second track obtained by a second sensor, the connection based on an association between the first track and the second track to represent a same object. The processor can determine, from a perspective of the tracks obtained by the first sensor compared to a perspective of the tracks obtained by the second sensor, a strength of the connection associated with each of the candidate track pairs. The processor can prune, responsive to identifying, based on the strength of the connection associated with each of the candidate track pairs, a connection with a weakest strength associated with the candidate track pairs, a weakest candidate track pair, the weakest candidate track pair having the connection with the weakest strength. The processor can also cluster, responsive to pruning the weakest candidate track pair, the candidate track pairs into at least a first cluster of the candidate track pairs and a second cluster of the candidate track pairs. The processor can determine, based on track-association techniques (e.g., pattern-matching techniques), a set of object track pairs from the at least first cluster of the can