Cost Matrix Input for Multiple Sensor Fusion

I have two sensors on a single platform (car) which are detecting a target (another car). Suppose

% Sensor 1 data
X1 = 100; Y1 = 10; Z1 = 0;
% Sensor @ data
X2 = 100.1; Y2 = 9.9; Z2 = 0;
% Convert them to object
obj_det = [objectDetection(0.5,[X1,Y1,Z1],'SensorIndex',1),objectDetection(1,[X2,Y2,Z2],'SensorIndex',2)];

Initate a tracker & set hasCostMatrixInput to false

tracker = trackerGNN();
tracker.Assignment = 'Auction';
tracker.ConfirmationThreshold = [2 5];
tracker.DeletionThreshold = [4 5];
tracker.AssignmentThreshold = [1000 inf];
tracker.HasCostMatrixInput = false;

Now if i call tracker function I have one track (which means that it fuses both sensors data)

[~,~,all_tracks,~] = tracker(obj_det,1)

all_tracks =

objectTrack with properties:

TrackID: 1

BranchID: 0

SourceIndex: 0

UpdateTime: 1

Age: 2

State: [6×1 double]

StateCovariance: [6×6 double]

StateParameters: [1×1 struct]

ObjectClassID: 0

TrackLogic: 'History'

TrackLogicState: [1 1 0 0 0]

IsConfirmed: 1

IsCoasted: 0

IsSelfReported: 1

ObjectAttributes: [1×1 struct]

My advisor's requirement is that i should give external costmatrix so i

Initate a tracker & set hasCostMatrixInput to true

tracker = trackerGNN();
tracker.Assignment = 'Auction';
tracker.ConfirmationThreshold = [2 5];
tracker.DeletionThreshold = [4 5];
tracker.AssignmentThreshold = [1000 inf];
tracker.HasCostMatrixInput = true;

Now for first call an empty matrix is required whose rows size matches number of detections so

cost_matrix_init_val = zeros(0,2);

Now if i call tracker function with cost matrix, i ve two seperate tracks

[~,~,all_tracks,~] = tracker(obj_det,1,cost_matrix_init_val)

all_tracks =

2×1 objectTrack array with properties:

TrackID

BranchID

SourceIndex

UpdateTime

Age

State

StateCovariance

StateParameters

ObjectClassID

TrackLogic

TrackLogicState

IsConfirmed

IsCoasted

IsSelfReported

ObjectAttributes

Question:

if i enabled has CostMatrixInput, how to do multisensor fusion in that case as tracker is not fusing them in that case?

Answer 1

[Elad Kivelevitch]

Sugar_daddy:

Unfortunately, when you set the HasCostMatrixInput to true, we can no longer make any assumptions on how to compute the distance between the two detections. In other words, we don't know what convention you use for the distance (Euclidean? Mahalanobis? Intersection-over-union? some other completely different distance?).

Even if we defaulted to use some distance measure, it would have a different "size" than the AssignmentThreshold you used. Compare Euclidean with Mahalanobis - two measurements can be 2 kms apart but if the measurement noise is 10 kms, they can actually have a very low Mahalanobis distance. Therefore, even if we computed the distance with some method, if we don't know what method you're using for AssignmentThreshold, it is meaningless.

So, we made the decision to not initialize multi-sensor tracks from two detections.

In the next call, they will be updated correctly because you'll pass their cost matrix.

Elad

[Sugar Daddy]

Thanks a lot for response Sir.