Transformer-based multi-target tracking (MTT) jointly learns data association and state estimation, but MT3/Track-MT3-style trackers repeatedly re-encode measurement windows, incurring redundant computation. We propose Causal Neural Set Filtering (CNSF)\footnote{Code: https://github.com/daihuangyu/CNSF}, a neural set filter that encodes only current measurements while carrying past evidence in a structured recursive track state. CNSF combines exclusive Sinkhorn association, association-conditioned Kalman-shaped updates with moment matching, and recurrent Bernoulli lifecycle modeling with measurement-driven birth. These mechanisms impose soft one-to-one constraints, propagate association-induced state uncertainty, and support existence estimation under missed detections and birth--death transitions. On a held-out three-regime simulated test set, CNSF reduces mean GOSPA and T-GOSPA relative to Track-MT3 by 19.3% and 30.4%, with 55.9% fewer parameters and a 3.76× speedup in single-thread CPU inference.
Figures & tables
Figure 1: Overview of CNSF. Top: causal single-frame inference via prediction, current-frame encoding, exclusive association, moment-matched update, and lifecycle/birth modeling. Bottom: training-only supervision, closed-loop curricula, and joint BPTT through the recursive path.
Figure 2: Fixed qualitative cases (run 000): three scenarios (rows) × five methods (columns), with identical measurements and ground truth. G is mean GOSPA over frames 20–99; axes are in meters.
Method
S1
S2
S3
G
P
T
CPU
Par.
MHT
.576
1.601
2.579
1.585
1.585 †
1.676
89.04
–
δ -GLMB
.250
.998
2.034
1.094
1.094 †
1.251
1614.57
–
PMBM
.270
.982
1.945
1.066
1.250
1.179
81.47
–
Track-MT3
.609
.990
1.680
1.093
1.245
1.443
117.37
19.38
Track-MT3-CM
.642
1.050
1.750
1.148
1.427
1.595
39.68
8.48
CNSF
.166
.828
1.652
.882
.990
1.004
31.18
8.54
Table 1: Held-out test performance. G/P/T : mean GOSPA/Pro-GOSPA/T-GOSPA; CPU: ms/frame; Par.: millions. † For deterministic point outputs with unit existence, Pro-GOSPA equals GOSPA.
Variant
Gˉ
Pˉ
Tˉ
M3
F3
Full CNSF
.882
.990
1.004
.759
.246
w/o excl. Sinkhorn
1.524
1.539
1.659
.724
1.325
w/o pairwise mix.
.931
1.053
1.060
.847
.250
w/o rec. lifecycle
1.112
1.194
1.248
1.195
.177
Table 2: Structural ablations on the held-out test set. Gˉ/Pˉ/Tˉ : mean GOSPA/Pro-GOSPA/T-GOSPA; M3/F3 : S3 missed/false.
We address the problem of searching for an unknown number of stationary targets at unknown positions with a mobile agent. A probability hypothesis density filter is used to estimate the expected number of targets under measurement uncertainty. Existing planners, such as Active Search (AS) and its Intermittent variant (ASI), achieve accurate detection but require costly online optimization. To reduce online computation, we propose to use a convolutional neural network to approximate AS or ASI decisions through direct inference. The network is trained on AS/ASI data using a multi-channel grid that encodes target beliefs, the agent position, visitation history, and boundary information. Simulations with uniform and clustered target distributions show that the network achieves detection rates comparable to AS or ASI while reducing computation by orders of magnitude.
Bilal Yousuf, Zsofia Lendek, Lucian Busoniu
Technical University of Cluj-Napoca, Romania · Corresponding Member of Romanian Academy
Multi-sensor tracking in the real world involves asynchronous sensors with partial coverage and heterogeneous detection performance. Although probabilistic tracking methods permit detection probability and clutter intensity to depend on state and sensing context, many practical frameworks enforce globally uniform observability assumptions. Under multi-rate and partially overlapping sensing, this simplification causes repeated non-detections from high-rate sensors to erode tracks visible only to low-rate sensors, potentially degrading fusion performance. We introduce DetectorContext, an abstraction for the open-source multi-target tracking framework Stone Soup. DetectorContext exposes detection probability and clutter intensity as state-dependent functions evaluated during hypothesis formation. The abstraction integrates with existing probabilistic trackers without modifying their update equations. Experiments on asynchronous radar-lidar data demonstrate that context-aware modeling restores stable fusion and significantly improves HOTA and GOSPA performance without increasing false tracks.
Martin Vonheim Larsen, Kim Mathiassen
Norwegian Defence Research Establishment (FFI) · University of Oslo, Department of Technology Systems
Multiobject tracking (MOT) is a fundamental capability in autonomous navigation, radar and sonar systems, as well as 6G integrated sensing-and-communication (ISAC) networks. In this paper, we propose a neural-enhanced track-before-detect (TkBD) framework that operates directly on complex-valued raw signals and learns a latent representation of unmodeled propagation effects during runtime. We introduce an environment-dependent latent intensity variable that characterizes the signal power of each object in the shared propagation environment. The probability density function of this intensity variable is represented by a normalizing flow and trained online in an unsupervised manner within the sequential Bayesian inference procedure. The learning objective is derived from a variational Bayes formulation by maximizing the evidence lower bound (ELBO), thereby accounting for the uncertainties inherent in the Bayesian TkBD problem. To enable scalable inference, we make use of a recently introduced Bayesian TkBD framework that combines belief propagation, which avoids high-dimensional operations, with sequential Monte Carlo methods for efficient low-dimensional updates.