Single Target Tracking Research Articles

An Underwater Single Target Tracking Method Using SiamRPN++ Based on Inverted Residual Bottleneck Block

Traditional tracking algorithms reply on manually extracted features while deep learning algorithms can automatically extract features, which is of great importance for single target tracking in the complex underwater environment that is characterized by poor visibility, low contrasts, and occlusion. The convolutional neural network SiamRPN++ is an advanced deep learning algorithm whose backbone network structure is. However, this algorithm is difficult to be implemented in an underwater platform due to its large amount of computation and high requirement for the computing power of the hardware. To solve this problem, a new backbone network structure NewNet-62 characterized by inverted residual bottleneck block and the SiamRPN++ algorithm are proposed. The depth-wise separable convolution in NewNet-62 simplifies the computation so that the SiamRPN++ algorithm can realize fast and accurate tracking of underwater single targets. The video set in VOT2016 was used to test the algorithm. The results show that the SiamRPN++ algorithm using NewNet-62 as the main network has the great performance. Compared with the algorithms using ResNet-50, the accuracy of the SiamRPN++ algorithm increased to 0.524 and its EAO increased to 0.303. And its tracking speed increased to 73.74. The network complexity reduced to 3.013 billion and the network parameters reduced to 12.538 million, which significantly reduces network complexity and number of network parameters.

Filtering in Pairwise Markov Model With Student's t Non-Stationary Noise With Application to Target Tracking

Hidden Markov models are widely used for target tracking, where the process and measurement noises are usually modeled as independent Gaussian distributions for mathematical simplicity. However, the independence and Gaussian assumptions do not always hold in practice. For example, in a typical target tracking application, a radar is utilized to track a non-cooperative target. Measurement noise is correlated over time since the sampling frequency of a radar is usually far greater than the bandwidth of measurement noise. Besides, when target is maneuvering, the process and measurement noises are heavy-tailed and non-Gaussian due to intrinsic data generation mechanism. In this paper, we consider a linear pairwise Markov model (PMM) with Student's t noise to model non-cooperative single target tracking without clutter and missed detections. A PMM is more general than an HMM and can be used to model correlated measurement noise or correlated process and measurement noises. The Student's t distribution is one of the most commonly used heavy-tailed distribution and can be used to address irregular target motion. We mainly focus on solving the filtering problems for the model. First, we develop a filter for the case where noise statistics are accurately known. Second, we further consider the case of non-stationary Student's t noise, and propose a novel robust filter by employing a variational Bayesian method. Finally, the effectiveness of the proposed filters is illustrated via simulation results.

Bayes-Optimal Set-Valued Tracking of Single Point Targets

Conventional single-point-target tracking algorithms are recursive point estimators with point-measurement input data. In less well-known approaches, the tracking algorithm is a recursive set estimator with point-measurement or set-measurement input data. This paper provides a very general, systematic, and theoretically rigorous foundation for Bayes-optimal single-target trackers of the latter kind; as well as specific trackers arising from that foundation. This foundation is intuitive and conceptually simple and, in particular, requires no measure-theoretic complexities.

Bernoulli Filters for Multiple Correlated Sensors

The Bernoulli filter is a general, Bayes-optimal solution for tracking a single disappearing and reappearing target, using a single sensor whose observations are corrupted by missed detections and a general, known clutter process. Like virtually all target-tracking algorithms it presumes a hidden Markov model (HMM) structure on the sensor and target. Pieczynski's pairwise Markov model (PMM) relaxes this assumption, thereby addressing correlated sensor noise and non-Markovian target motion. In an earlier paper, we derived a “PMM Bernoulli filter” that obeys PMM rather than restrictive HMM sensor/target statistics. This paper generalizes both the HMM and PMM Bernoulli filters to the case of multiple, possibly correlated sensors, resulting in a general, Bayes-optimal single-target tracker for complexly correlated sensors.

Distributed Kalman Filtering Based on the Non-Repeated Diffusion Strategy.

Estimation accuracy is the core performance index of sensor networks. In this study, a kind of distributed Kalman filter based on the non-repeated diffusion strategy is proposed in order to improve the estimation accuracy of sensor networks. The algorithm is applied to the state estimation of distributed sensor networks. In this sensor network, each node only exchanges information with adjacent nodes. Compared with existing diffusion-based distributed Kalman filters, the algorithm in this study improves the estimation accuracy of the networks. Meanwhile, a single-target tracking simulation is performed to analyze and verify the performance of the algorithm. Finally, by discussion, it is proved that the algorithm exhibits good all-round performance, not only regarding estimation accuracy.

Adapting to the motion of multiple independent targets using multileaf collimator tracking for locally advanced prostate cancer: Proof of principle simulation study.

For patients with locally advanced cancer, multiple targets are treated simultaneously with radiotherapy. Differential motion between targets can compromise the treatment accuracy, yet there are currently no methods able to adapt to independent target motion. This study developed a multileaf collimator (MLC) tracking algorithm for differential motion adaptation and evaluated it in simulated treatments of locally advanced prostate cancer. A multi-target MLC tracking algorithm was developed that consisted of three steps: (a) dividing the MLC aperture into two possibly overlapping sections assigned to the prostate and lymph nodes, (b) calculating the ideally shaped MLC aperture as a union of the individually translated sections, and (c) fitting the MLC positions to the ideal aperture shape within the physical constraints of the MLC leaves. The multi-target tracking method was evaluated and compared with two existing motion management methods: single-target tracking and no tracking. Treatment simulations of six locally advanced prostate cancer patients with three prostate motion traces were performed for all three motion adaptation methods. The geometric error for each motion adaptation method was calculated using the area of overexposure and underexposure of each field. The dosimetric error was estimated by calculating the dose delivered to the prostate, lymph nodes, bladder, rectum, and small bowel using a motion-encoded dose reconstruction method. Multi-target MLC tracking showed an average improvement in geometric error of 84% compared to single-target tracking, and 83% compared to no tracking. Multi-target tracking maintained dose coverage to the prostate clinical target volume (CTV) D98% and planning target volume (PTV) D95% to within 4.8% and 3.9% of the planned values, compared to 1.4% and 0.7% with single-target tracking, and 20.4% and 31.8% with no tracking. With multi-target tracking, the node CTV D95%, PTV D90%, and gross tumor volume (GTV) D95% were within 0.3%, 0.6%, and 0.3% of the planned values, compared to 9.1%, 11.2%, and 21.1% for single-target tracking, and 0.8%, 2.0%, and 3.2% with no tracking. The small bowel V57% was maintained within 0.2% to the plan using multi-target tracking, compared to 8% and 3.5% for single-target tracking and no tracking, respectively. Meanwhile, the bladder and rectum V50% increased by up to 13.6% and 5.2%, respectively, using multi-target tracking, compared to 2.7% and 1.9% for single-target tracking, and 11.2% and 11.5% for no tracking. A multi-target tracking algorithm was developed and tracked the prostate and lymph nodes independently during simulated treatments. As the algorithm optimizes for target coverage, tracking both targets simultaneously may increase the dose delivered to the organs at risk.

Automated extraction of dolphin whistles-A sequential Monte Carlo probability hypothesis density approach.

The need for automated methods to detect and extract marine mammal vocalizations from acoustic data has increased in the last few decades due to the increased availability of long-term recording systems. Automated dolphin whistle extraction represents a challenging problem due to the time-varying number of overlapping whistles present in, potentially, noisy recordings. Typical methods utilize image processing techniques or single target tracking, but often result in fragmentation of whistle contours and/or partial whistle detection. This study casts the problem into a more general statistical multi-target tracking framework and uses the probability hypothesis density filter as a practical approximation to the optimal Bayesian multi-target filter. In particular, a particle version, referred to as a sequential Monte Carlo probability hypothesis density (SMC-PHD) filter, is adapted for frequency tracking and specific models are developed for this application. Based on these models, two versions of the SMC-PHD filter are proposed and the performance of these versions is investigated on an extensive real-world dataset of dolphin acoustic recordings. The proposed filters are shown to be efficient tools for automated extraction of whistles, suitable for real-time implementation.

Decorrelation of Previously Communicated Information for an Interacting Multiple Model Filter

In a sensor network compensation of the correlated information caused by previous communication is of utmost interest for distributed estimation. In this article, we investigate an information decorrelation approach that can be applied when using interacting multiple model filters in the sensor nodes for a family of jump Markov linear systems. Implementation issues that might arise while applying the decorrelation approach are addressed in detail. The investigated approach is compared with alternatives on simple distributed single maneuvering target tracking examples.

Single target tracking via correlation filter and context adaptively

Recently, tracking methods based on correlation filter show impressive performance in a variety of complex environments for their excellent classification performance. However, most of the existing methods only focus on the information in the bounding box regions of the target, and do not fully utilized the contextual information and the internal structure information of the target. Thus, when the target scenes experience dramatic change, the learned filter could not accurately adapt to the appearance change, which will lead to model degradation of the target. To address this issue, a novel approach via correlation filter and context jointly is proposed in this paper. First, we decompose the target into multiple independent parts, and each part learns the filtering response separately. Through the joint learning of multiple independent filters, the target model can effectively maintain the structural information of the object and is not sensitive to partial occlusion, and etc. Second, we introduce multi-channel features in the representation based on the parts of the target and the contextual information to migrate the background influences. With the introduction of collaborative representation strategy, the impact of background noise can be effectively suppressed. To evaluate the proposed approach, we conduct extensive experiments on several challenging benchmark datasets including OTB-2013 and OTB-2015 datasets. The results show our method demonstrates comparable performance against several state-of-the-art methods.

Single Object Tracking for Dynamic Programming based on Fine-grained Network

Single target tracking has always been a key and challenging research field in computer vision. Currently, an increasing number of researchers are focusing on extracting better tracking features and designing the best tracker. This paper proposes a new single target tracking network that uses fine-grained features and dynamic programming (DPFNet). In order to extract superior features, we added an attention module to the regression network enabling us to extract finer-grained and discriminative features to achieve regression. Besides, we did observe that different objects have varying moving rates; for different moving targets, the magnitude of the changes in target position within two adjacent frames is not the same either. Although an area search of 4 times the target’s size can be applied to most objects, targets with large position changes may appear in other image areas outside the search area and the target would not be located as a result. Aiming at solving this problem, when designing the tracker, this paper analyzes some of the indicators for predicting the location and uses the analysis results to determine whether the search area is appropriate, so as to dynamically adjust the extent of the search area thereby significantly improving the tracking function. In other words, the size of the search area can be dynamically recalibrated for different images. Subsequent experiments prove that the method put forward in this paper achieves State-of-the-Art results.

Optical Flow in Deep Visual Tracking

Single-target tracking of generic objects is a difficult task since a trained tracker is given information present only in the first frame of a video. In recent years, increasingly many trackers have been based on deep neural networks that learn generic features relevant for tracking. This paper argues that deep architectures are often fit to learn implicit representations of optical flow. Optical flow is intuitively useful for tracking, but most deep trackers must learn it implicitly. This paper is among the first to study the role of optical flow in deep visual tracking. The architecture of a typical tracker is modified to reveal the presence of implicit representations of optical flow and to assess the effect of using the flow information more explicitly. The results show that the considered network learns implicitly an effective representation of optical flow. The implicit representation can be replaced by an explicit flow input without a notable effect on performance. Using the implicit and explicit representations at the same time does not improve tracking accuracy. The explicit flow input could allow constructing lighter networks for tracking.

Maneuvering Target Tracking Based on Multiple-model Multipath Bernoulli Filter

Aiming at the problem of maneuvering single target tracking in over the horizon radar system under the clutter environment, a novel multiple-model multipath Bernoulli filter (MM-MPBF) algorithm is proposed. First, the multipath target measurement model is established based on the finite set statistics, and then combining the multiple models method with the multipath Bernoulli filter to deal with the single maneuvering target tracking problem in over the horizon radar system. Finally, the simulation experiment is used to verify the performance of the MM-MPBF algorithm.

UAV monitoring and forecasting model in intelligent traffic oriented applications

Intelligent transportation system is a traffic management system developed with the progress of society and traffic. Its idea is to integrate the real-time operation of people, vehicles, roads and traffic involved in the traffic. The purpose of this paper is to build a safe, reliable and efficient vehicle monitoring and forecasting model for IOT. Based on the Beidou satellite positioning technology and Lora communication technology, aiming at the problem that the deep learning detection method cannot meet the real-time requirements in processing the monitoring video, this paper proposes a method of using multiple single target trackers instead of some yolov3 detection tasks, and puts forward the design idea and specific implementation scheme of the vehicle monitoring and prediction model. The vehicle monitoring and prediction model is used to detect four kinds of targets, namely, small cars, buses, trucks and pedestrians. The multi-target trajectory tracking is used to carry out the traffic statistics of multi vehicle types, the detection of two kinds of abnormal behaviors of traffic targets is low speed and parking, and the capture of pedestrians. The experimental results show that the vehicle monitoring and prediction model has the highest accuracy of location and type recognition for four types of traffic objects, namely, small cars, trucks, buses and pedestrians, reaching 80%.

Attitude estimation of a permanent magnet spherical motor based on an improved fast discriminative scale space tracking algorithm

A permanent magnet spherical motor (PMSM) can perform a three-degree-of-freedom operation. The estimation of the attitude of the rotor is one of the key steps for realizing the closed-loop control of PMSM. An attitude estimation method based on a target-tracking algorithm is proposed in this study. Firstly, a high-speed camera is used to capture a video sequence of the spherical motor in motion. The location of the target is obtained by introducing an improved fast discriminative scale space tracking (IFDSST) algorithm to track the video sequence. In consideration of the motor’s output shaft being covered with load, the single-target tracking algorithm is improved into a multi-target tracking algorithm. Markers, which are symmetrical with respect to the output shaft, are made on the surface of the rotor, and their attitudes are obtained from the images captured by the tracking algorithm. Secondly, the attitude of the output shaft is estimated by the coordinate transformation and the positional relationship between the markers and the output shaft. The effectiveness and superiority of the proposed algorithm are verified by comparing it with a micro-electromechanical system (MEMS) method and a fast -complementary filter (FCF) method. Three types of motor motions, including yaw, spin, and tilt, are tested in the experiments, and the average angle error or trajectory error of the proposed IFDSST method reduces around 33%, 53%, and 79% as compared with those of the MEMS and FCF methods, respectively. Experiments show that the proposed algorithm can effectively and accurately measure the attitude of the motor’s rotor.

Auxiliary-Filter-Free Incompressible Particle Flow Filtering Using Direct Estimation of the Log-Density Gradient with Target Tracking Examples

This paper presents an incompressible particle flow filtering method that does not require an auxiliary filter by estimating log-density gradients directly from particles. Particle flow filter (PFF) is likely to avoid particle impoverishment and degeneracy problems that occur in particle filters because particles themselves move toward desired density to perform measurement updates. There are various implementation forms for PFF depending on the assumptions made about the flow. This paper deals with PFF using incompressible flow. Incompressible PFF requires the log-density gradient to calculate the flow. The well-known gradient estimation method for incompressible PFF is a finite difference method collaborating with k nearest neighbors(kNN) method. Since this method requires the prior knowledge about the prior density value in each particle, it is necessary to use an auxiliary filter or a density estimation technique. As a result, the performance of an auxiliary filter or a density estimation technique can directly affect the PFF performance, and the finite difference method is more likely to be inaccurate than directly estimating the log-density gradient. Therefore, this paper presents a PFF structure applying least-squares log-density gradient (LSLDG) method that estimates the log-density gradient directly from particles. In order to verify the performance of the presented structure, this paper performs both single and multiple target tracking simulations. Simulation results demonstrate that the presented structure has a relatively good estimation performance and works more robustly for various situations.

Distributed Sensor Management Based on Target Losing Probability for Maneuvering Multi-Target Tracking

This paper proposed a distributed sensor management method in the constraint of target losing probability for maneuvering multi-target tracking. Three main contributions existed in this paper, including adding the emission interception risk to the sensor management objective function, analyzing actual working modes of radars and regarding the target losing probability as one of constraints in sensor management. Two kind of radar working modes were introduced. They were the single target tracking mode and the multiple target tracking mode, leading to different kinds of sensor management methods. Then two sensor management models, containing sensor management in a large defending scope and in a small defending scope, were built. After that, the distributed sensor management process was introduced. Finally, simulations were conducted to show the effectiveness and efficiency of the proposed sensor management methods. By applying the proposed sensor management method, targets couldn't be easily lost and radar could be well protected at the same time.

Multisensor Multi-Target Tracking Based on GM-PHD Using Out-Of-Sequence Measurements.

In this paper, we study the issue of out-of-sequence measurement (OOSM) in a multi-target scenario to improve tracking performance. The OOSM is very common in tracking systems, and it would result in performance degradation if we used it inappropriately. Thus, OOSM should be fully utilized as far as possible. To improve the performance of the tracking system and use OOSM sufficiently, firstly, the problem of OOSM is formulated. Then the classical B1 algorithm for OOSM problem of single target tracking is given. Next, the random finite set (RFS)-based Gaussian mixture probability hypothesis density (GM-PHD) is introduced. Consequently, we derived the equation for re-updating of posterior intensity with OOSM. Implementation of GM-PHD using OOSM is also given. Finally, several simulations are given, and results show that tracking performance of GM-PHD using OOSM is better than GM-PHD using in-sequence measurement (ISM), which can strongly demonstrate the effectiveness of our proposed algorithm.

An Efficient Multi-Path Multitarget Tracking Algorithm for Over-The-Horizon Radar.

In target tracking environments using over-the-horizon radar (OTHR), one target may generate multiple detections through different signal propagation paths. Trackers need to jointly handle the uncertainties stemming from both measurement origin and measurement path. Traditional multitarget tracking algorithms suffer from high computational loads in such environments since they need to enumerate all possible joint measurement-to-track assignments considering the measurements paths unless they employ some approximations regarding the measurements and their corresponding paths. In this paper, we propose a novel algorithm, named multi-path linear multitarget integrated probabilistic data association (MP-LM-IPDA), to efficiently track multitarget in multiple detection environments. Instead of generating all possible joint assignments, MP-LM-IPDA calculates the modulated clutter measurement density for each measurement cell of each track. The modulated clutter measurement density considers the possibility that the measurement cells originate from the clutter as well as from other potential targets. By incorporating the modulated clutter measurement density, the single target tracking structure can be applied for multitarget tracking, which significantly reduces the computational load. The simulation results demonstrate the effectiveness and efficiency of the proposed algorithm.

MVDR code receiver for code multiplexing MIMO radar

Multiple-input multiple-output (MIMO) radar recently has attracted the attention of engineers and researchers. MIMO radar performance is based on the orthogonality of the transmitted waveforms. A receiver optimisation method is proposed to suppress cross-correlation rather than finding orthogonal waveforms. The method considers two transmitters and is based on the minimum variance distorsionless response (MVDR) algorithm to minimise cross-correlation while maintaining the level of auto-correlation. Performance is evaluated by simulation using random phase code sequences. The proposed method is well suited for single target tracking, or slow moving target detection.

Hierarchical fully adaptive radar

By emulating the cognitive perception-action cycle believed to be at the core of animal cognition, cognitive radars promise to improve radar performance over standard systems. The fully adaptive radar (FAR) framework provides a generalised approach to implementing a single cognitive perception-action cycle for radar systems, but complex adaptive problems necessitate the interaction of multiple perception-action cycles. This study describes the general form of the hierarchical FAR (HFAR) framework. The HFAR framework is applied to a single-target tracking, sensor fusion problem, and real-time experimental results demonstrate the efficacy of the proposed architecture for handling problems of varying scales in a consistent, adaptive fashion.