Feng Yang

In this paper, a distributed cubature joint probabilistic data association (JPDA) filter is proposed for circular formation target tracking under nonlinear multi-sensor tracking system. First, a structure identification and constraint algorithm for two types of circular formation groups is utilized to modify estimation results from cubature JPDA filter. Then, after achieving multi-sensor data association through a two-tiered algorithm proposed in this paper, a hybrid consensus-based fusion strategy, which combines the complementary features of consensus on measurement and consensus on information, is performed on cubature JPDA filter, that extensively exploits the information spread through the network and improves the tracking accuracy. Subsequently, this paper proves the stability of proposed tracking algorithm. Simulations are presented to verify the effectiveness of the proposed algorithm.

Transformer-based multiobject tracking (MOT) methods have attracted significant attention from researchers. However, these Transformer-based models often suffer from suboptimal inference speeds due to their architectural complexities or other inherent issues, rendering them difficult to deploy in practical industrial applications. To address this challenge, we revisited the classic joint detection and tracking (JDT) paradigm and analyzed existing models. Drawing inspiration from Detection Transformer’s (DETR) object queries, which naturally encode object appearance features, we constructed a fast and novel JDT-type MOT framework named FastTrackTr by implementing an efficient interframe information transfer mechanism. This framework integrates three key technological innovations: a cross-decoder mechanism that implicitly incorporates historical trajectory information without requiring additional queries or decoders, a historical encoder and decoder pair for refining and utilizing historical feature representations, and a deterministic fixed-shape architecture that enables seamless TensorRT acceleration. Benefiting from these designs, our approach not only reduces the number of queries required for tracking but also avoids introducing excessive network structures, ensuring model simplicity while maintaining high accuracy. Experimental results show that our method achieves real-time tracking while maintaining state-of-the-art accuracy. On an NVIDIA RTX 4090 with an image size of 1333 × 800, it reaches 62.4 higher order tracking accuracy (HOTA) at 86.6 frames per second (FPS) on the DanceTrack dataset, outperforming other advanced Transformer-based methods. Furthermore, it excels on edge devices, such as the NVIDIA Jetson AGX Orin, where its lightweight variant achieves up to 59.2 FPS on 640 × 640 images, meeting real-time requirements for practical applications.

This paper addresses the issue of multi-target tracking in a distributed sensor network, where sensors have different fields of view (FoVs) and the scenario is characterized by multiple uncertainties on detection probability, clutter rate and noise covariances. First, we design a robust PHD (RPHD) filter, based on a target augmented state and a clutter feature state, to jointly estimate the target cardinality and kinematic states as well as multiple unknown parameters (noise covariances, detection probability and clutter rate). Gaussian Inverse Gamma Inverse Wishart Mixture (GIGIWM) and inverse Gamma mixture representations are used for the intensities of target augmented state and clutter state, respectively, and the estimation of all the unknown variables is achieved by means of variational Bayesian inference incorporating constrained distributions. Secondly, for a distributed sensor network, we propose a distributed fusion algorithm that extends the weighted arithmetic average (WAA) strategy to fuse information within overlapping and non-overlapping regions and to deal with local posteriors containing both global and local variables. Lastly, the local information confidence of each sensor posterior is incorporated into the fusion weight design to enhance the effective utilization of information in fusion. Numerical simulation experiments are performed to demonstrate the superior performance and strong robustness of the presented filter and distributed fusion algorithm.

Vision-based multi-view perception systems, especially those using bird’s-eye view (BEV) representations, have become increasingly important in autonomous driving. Current state-of-the-art methods typically transform image features from multiple camera views into the BEV space via explicit or implicit depth estimation. However, they often rely on uniform or fixed-height sampling and lack height-aware priors, making them insensitive to the fact that different object categories occupy distinct local height ranges. Furthermore, most approaches treat BEV features as independent across grid locations, overlooking the structured correlations between different parts of an object in 3D space. These limitations hinder accurate spatial reasoning in both vertical and horizontal dimensions. To address these issues, we propose HV-BEV, a novel BEV perception framework that decouples the feature sampling process into Horizontal feature aggregation and Vertical adaptive height-aware reference point sampling. Specifically, for horizontal modeling, we dynamically construct a set of relevant neighboring points on the ground-aligned plane for each 3D reference point, facilitating structured cross-view feature aggregation and promoting consistent representation of large or partially visible objects. For vertical modeling, we introduce an adaptive height-aware module that leverages historical information to guide 3D reference points to focus on the plausible height regions where objects of interest are likely to appear, replacing fixed uniform height sampling. Extensive experiments on the nuScenes dataset demonstrate the effectiveness of our method. Our HV-BEV framework consistently outperforms baselines, achieving 50.5% mAP and 59.8% NDS on the nuScenes test set. Code is available at https://github.com/Uddd821/HV-BEV

This paper addresses communication bandwidth reduction and energy efficiency enhancement of a peer-to-peer sensor network for distributed multitarget tracking under unknown clutter rate and detection profile. We propose an event-triggered distributed bootstrap cardinalized probability hypothesis density (CPHD) filter, where the local filter consists of an estimator that estimates the clutter and detection parameters, and a tracker that estimates the number and states of targets. Specifically, the proposed approach enforces each sensor to broadcast its local posterior density only when it has high estimation quality and significant information gain, wherein the former is determined by the outputs of the local estimator and tracker, while the latter is measured via the Jensen-Shannondivergence. Then, we develop a modified weighted arithmetic average fusion method to separately fuse the cardinality distributions and intensity functions, in which the fusion weights are constructed based on estimation qualities and triggering thresholds, which is incorporated into the distributed flooding scheme for distributed averaging. Additionally, the Gaussian mixture implementations of the proposed event-triggered distributed bootstrap CPHD filter are given, where only Gaussian components with high weights are applied, thus reducing the effect of false alarms and, consequently, achieving a partial consensus across the network. Finally, the effectiveness of the proposed algorithm is validated through simulation experiments.

The burgeoning demand for unmanned aerial vehicles (UAVs) across diverse domains can be attributed to their high flexibility, ease of deployment and low operational costs. Concomitantly, rapid advancements in reinforcement learning have emerged as a viable avenue for augmenting the autonomy of UAVs. This paper provides a comprehensive overview of the foundational concepts and methodologies of reinforcement learning and taxonomizes its applications in UAV decision-making into three primary categories: fundamental tasks encompassing obstacle avoidance and path planning, advanced tasks involving cooperative control, and complex tasks requiring adversarial decision-making. Additionally, the challenges associated with implementing reinforcement learning in UAV applications are critically examined. In the final section, we envision future research directions and provide a comprehensive summary of the study. This will assist practitioners and researchers in selecting appropriate reinforcement learning algorithms for their drone mission applications.

Multi-Object Tracking (MOT) is a critical problem in computer vision, yet current research on tracking the motion state of objects relative to the ground remains limited. The extant literature exhibits a notable dearth in the exploration of this aspect. Deep learning methodologies encounter challenges in accurately discerning object motion states, while conventional approaches reliant on comprehensive mathematical modeling may yield sub-optimal tracking accuracy. To address these challenges, we introduce a Model-Data-Driven Motion State Judgment Object Tracking (MoD2T) method. This innovative architecture adeptly amalgamates classical mathematical modeling with deep learning-based MOT frameworks. The integration of mathematical modeling and deep learning within MoD2T enhances the precision of object motion state determination, thereby elevating tracking accuracy. Our empirical investigations comprehensively validate the efficacy of MoD2T across varied scenarios, encompassing unmanned aerial vehicle surveillance and street-level tracking. Furthermore, to gauge the method’s adeptness in discerning object motion states, we introduce the Motion State Validation F1 (MVF1) metric.