-
Visual Multi-Object Tracking with Re-Identification and Occlusion Handling using Labeled Random Finite Sets
Authors:
Linh Van Ma,
Tran Thien Dat Nguyen,
Changbeom Shim,
Du Yong Kim,
Namkoo Ha,
Moongu Jeon
Abstract:
This paper proposes an online visual multi-object tracking (MOT) algorithm that resolves object appearance-reappearance and occlusion. Our solution is based on the labeled random finite set (LRFS) filtering approach, which in principle, addresses disappearance, appearance, reappearance, and occlusion via a single Bayesian recursion. However, in practice, existing numerical approximations cause rea…
▽ More
This paper proposes an online visual multi-object tracking (MOT) algorithm that resolves object appearance-reappearance and occlusion. Our solution is based on the labeled random finite set (LRFS) filtering approach, which in principle, addresses disappearance, appearance, reappearance, and occlusion via a single Bayesian recursion. However, in practice, existing numerical approximations cause reappearing objects to be initialized as new tracks, especially after long periods of being undetected. In occlusion handling, the filter's efficacy is dictated by trade-offs between the sophistication of the occlusion model and computational demand. Our contribution is a novel modeling method that exploits object features to address reappearing objects whilst maintaining a linear complexity in the number of detections. Moreover, to improve the filter's occlusion handling, we propose a fuzzy detection model that takes into consideration the overlapping areas between tracks and their sizes. We also develop a fast version of the filter to further reduce the computational time.
△ Less
Submitted 11 July, 2024;
originally announced July 2024.
-
General-Purpose Retrieval-Enhanced Medical Prediction Model Using Near-Infinite History
Authors:
Junu Kim,
Chaeeun Shim,
Bosco Seong Kyu Yang,
Chami Im,
Sung Yoon Lim,
Han-Gil Jeong,
Edward Choi
Abstract:
Machine learning (ML) has recently shown promising results in medical predictions using electronic health records (EHRs). However, since ML models typically have a limited capability in terms of input sizes, selecting specific medical events from EHRs for use as input is necessary. This selection process, often relying on expert opinion, can cause bottlenecks in development. We propose Retrieval-E…
▽ More
Machine learning (ML) has recently shown promising results in medical predictions using electronic health records (EHRs). However, since ML models typically have a limited capability in terms of input sizes, selecting specific medical events from EHRs for use as input is necessary. This selection process, often relying on expert opinion, can cause bottlenecks in development. We propose Retrieval-Enhanced Medical prediction model (REMed) to address such challenges. REMed can essentially evaluate unlimited medical events, select the relevant ones, and make predictions. This allows for an unrestricted input size, eliminating the need for manual event selection. We verified these properties through experiments involving 27 clinical prediction tasks across four independent cohorts, where REMed outperformed the baselines. Notably, we found that the preferences of REMed align closely with those of medical experts. We expect our approach to significantly expedite the development of EHR prediction models by minimizing clinicians' need for manual involvement.
△ Less
Submitted 22 July, 2024; v1 submitted 31 October, 2023;
originally announced October 2023.
-
Label Space Partition Selection for Multi-Object Tracking Using Two-Layer Partitioning
Authors:
Ji Youn Lee,
Changbeom Shim,
Hoa Van Nguyen,
Tran Thien Dat Nguyen,
Hyunjin Choi,
Youngho Kim
Abstract:
Estimating the trajectories of multi-objects poses a significant challenge due to data association ambiguity, which leads to a substantial increase in computational requirements. To address such problems, a divide-and-conquer manner has been employed with parallel computation. In this strategy, distinguished objects that have unique labels are grouped based on their statistical dependencies, the i…
▽ More
Estimating the trajectories of multi-objects poses a significant challenge due to data association ambiguity, which leads to a substantial increase in computational requirements. To address such problems, a divide-and-conquer manner has been employed with parallel computation. In this strategy, distinguished objects that have unique labels are grouped based on their statistical dependencies, the intersection of predicted measurements. Several geometry approaches have been used for label grouping since finding all intersected label pairs is clearly infeasible for large-scale tracking problems. This paper proposes an efficient implementation of label grouping for label-partitioned generalized labeled multi-Bernoulli filter framework using a secondary partitioning technique. This allows for parallel computation in the label graph indexing step, avoiding generating and eliminating duplicate comparisons. Additionally, we compare the performance of the proposed technique with several efficient spatial searching algorithms. The results demonstrate the superior performance of the proposed approach on large-scale data sets, enabling scalable trajectory estimation.
△ Less
Submitted 22 October, 2023;
originally announced October 2023.
-
A Moving Window Based Approach to Multi-scan Multi-Target Tracking
Authors:
Diluka Moratuwage,
Changbeom Shim,
Yuthika Punchihewa
Abstract:
Multi-target state estimation refers to estimating the number of targets and their trajectories in a surveillance area using measurements contaminated with noise and clutter. In the Bayesian paradigm, the most common approach to multi-target estimation is by recursively propagating the multi-target filtering density, updating it with current measurements set at each timestep. In comparison, multi-…
▽ More
Multi-target state estimation refers to estimating the number of targets and their trajectories in a surveillance area using measurements contaminated with noise and clutter. In the Bayesian paradigm, the most common approach to multi-target estimation is by recursively propagating the multi-target filtering density, updating it with current measurements set at each timestep. In comparison, multi-target smoothing uses all measurements up to current timestep and recursively propagates the entire history of multi-target state using the multi-target posterior density. The recent Generalized Labeled Multi-Bernoulli (GLMB) smoother is an analytic recursion that propagate the labeled multi-object posterior by recursively updating labels to measurement association maps from the beginning to current timestep. In this paper, we propose a moving window based solution for multi-target tracking using the GLMB smoother, so that only those association maps in a window (consisting of latest maps) get updated, resulting in an efficient approximate solution suitable for practical implementations.
△ Less
Submitted 8 October, 2022;
originally announced October 2022.
-
Multi-Scan Multi-Sensor Multi-Object State Estimation
Authors:
D. Moratuwage,
B. -N. Vo,
B. -T. Vo,
C. Shim
Abstract:
If computational tractability were not an issue, multi-object estimation should integrate all measurements from multiple sensors across multiple scans. In this article, we propose an efficient numerical solution to the multi-scan multi-sensor multi-object estimation problem by computing the (labeled) multi-sensor multi-object posterior density. Minimizing the $L_{1}$-norm error from the exact post…
▽ More
If computational tractability were not an issue, multi-object estimation should integrate all measurements from multiple sensors across multiple scans. In this article, we propose an efficient numerical solution to the multi-scan multi-sensor multi-object estimation problem by computing the (labeled) multi-sensor multi-object posterior density. Minimizing the $L_{1}$-norm error from the exact posterior density requires solving large-scale multi-dimensional assignment problems that are NP-hard. An efficient multi-dimensional assignment algorithm is developed based on Gibbs sampling, together with convergence analysis. The resulting multi-scan multi-sensor multi-object estimation algorithm can be applied either offline in one batch or recursively. The efficacy of the algorithm is demonstrated using numerical experiments with a simulated dataset.
△ Less
Submitted 2 December, 2022; v1 submitted 30 May, 2022;
originally announced May 2022.
