Tracking Dataset Research Articles

Multiple Traffic Target Tracking with Spatial-Temporal Affinity Network.

Traffic target tracking is a core task in intelligent transportation system because it is useful for scene understanding and vehicle autonomous driving. Most state-of-the-art (SOTA) multiple object tracking (MOT) methods adopt a two-step procedure: object detection followed by data association. The object detection has made great progress with the development of deep learning. However, the data association still heavily depends on hand crafted constraints, such as appearance, shape, and motion, which need to be elaborately trained for a special object. In this study, a spatial-temporal encoder-decoder affinity network is proposed for multiple traffic targets tracking, aiming to utilize the power of deep learning to learn a robust spatial-temporal affinity feature of the detections and tracklets for data association. The proposed spatial-temporal affinity network contains a two-stage transformer encoder module to encode the features of the detections and the tracked targets at the image level and the tracklet level, aiming to capture the spatial correlation and temporal history information. Then, a spatial transformer decoder module is designed to compute the association affinity, where the results from the two-stage transformer encoder module are fed back to fully capture and encode the spatial and temporal information from the detections and the tracklets of the tracked targets. Thus, efficient affinity computation can be applied to perform data association in online tracking. To validate the effectiveness of the proposed method, three popular multiple traffic target tracking datasets, KITTI, UA-DETRAC, and VisDrone, are used for evaluation. On the KITTI dataset, the proposed method is compared with 15 SOTA methods and achieves 86.9% multiple object tracking accuracy (MOTA) and 85.71% multiple object tracking precision (MOTP). On the UA-DETRAC dataset, 12 SOTA methods are used to compare with the proposed method, and the proposed method achieves 20.82% MOTA and 35.65% MOTP, respectively. On the VisDrone dataset, the proposed method is compared with 10 SOTA trackers and achieves 40.5% MOTA and 74.1% MOTP, respectively. All those experimental results show that the proposed method is competitive to the state-of-the-art methods by obtaining superior tracking performance.

Evaluating Automatic Body Orientation Detection for Indoor Location from Skeleton Tracking Data to Detect Socially Occupied Spaces Using the Kinect v2, Azure Kinect and Zed 2i.

Analysing the dynamics in social interactions in indoor spaces entails evaluating spatial–temporal variables from the event, such as location and time. Additionally, social interactions include invisible spaces that we unconsciously acknowledge due to social constraints, e.g., space between people having a conversation with each other. Nevertheless, current sensor arrays focus on detecting the physically occupied spaces from social interactions, i.e., areas inhabited by physically measurable objects. Our goal is to detect the socially occupied spaces, i.e., spaces not physically occupied by subjects and objects but inhabited by the interaction they sustain. We evaluate the social representation of the space structure between two or more active participants, so-called F-Formation for small gatherings. We propose calculating body orientation and location from skeleton joint data sets by integrating depth cameras. The body orientation is derived by integrating the shoulders and spine joint data with head/face rotation data and spatial–temporal information from trajectories. From the physically occupied measurements, we can detect socially occupied spaces. In our user study implementing the system, we compared the capabilities and skeleton tracking datasets from three depth camera sensors, the Kinect v2, Azure Kinect, and Zed 2i. We collected 32 walking patterns for individual and dyad configurations and evaluated the system’s accuracy regarding the intended and socially accepted orientations. Experimental results show accuracy above 90% for the Kinect v2, 96% for the Azure Kinect, and 89% for the Zed 2i for assessing socially relevant body orientation. Our algorithm contributes to the anonymous and automated assessment of socially occupied spaces. The depth sensor system is promising in detecting more complex social structures. These findings impact research areas that study group interactions within complex indoor settings.

Variability in Foraging Range and Direction Among Colonies in a Widespread Seabird, the Magellanic Penguin

Most seabirds forage far from land, making them hard to observe when foraging. Satellite tracking of seabirds shows where they come into conflict with human uses of the ocean, and whether they use protected areas. Because tracking data are expensive, data from one colony and/or year are sometimes used to design marine protection for a species across its range. Two assumptions commonly made are that foraging distance increases with colony size and individuals are uniformly distributed around colonies. We tested these assumptions using Magellanic penguinsSpheniscus magellanicusas an example. We used a large tracking dataset of 338 penguins foraging for chicks at 10 colonies in Argentina from 1996 to 2019. Foraging distance increased with population size among colonies, but predicted distances would not cover foraging areas for all colonies. There was no relationship between population size and foraging distance within colony among years for colonies with ten and 23 years of data. Penguins were not uniformly distributed around colonies. Penguins used ~24% (12-40%) of the ocean available within the colony’s maximum foraging distance. We also show that overlap between penguin foraging areas and marine protected areas (MPA) and hydrocarbon concessions varied among colonies partly because of variation in how far offshore penguins forage. Overlap with MPAs was low (0% – 20%) for seven of the ten colonies and high (23% – 100%) for the other three. Overlap with a large area permitted for hydrocarbon exploration (seismic surveys) was relatively high (23% – 81%) for seven colonies where penguins forage offshore. Data from one colony are unlikely to indicate the most effective marine spatial planning for all colonies. Our data show that to be effective, marine planning should consider the temporal and spatial dynamics of ocean conditions and the response of marine wildlife to these changes. Climate variability is predicted to increase, making knowledge of foraging-location variation among colonies and years critical to conservation planning.

Sensitivity of typhoon wind hazard in coastal region to the track modelling and the considered historical best track database

The tropical cyclone (TC) wind hazard is often assessed using the autoregressive type of model and the beta-advection model. In the present study, the beta-advection models are developed using the best track datasets from the China Meteorological Administration (CMA) and the Joint Typhoon Warning Center (JTWC). A comparison of statistics of the TC track parameters along the coastline of mainland China is presented using historical tracks and simulated tracks from the developed models. The comparison is extended by considering an existing autoregressive type of track model. Results show that the number of genesis per year based on the dataset in CMA is about 7% greater than that in JTWC. A comparison of the estimated T-year return period value of the annual maximum TC wind speed, vA-T, is presented using the three different track models. vA-T obtained based on the beta-advection model developed using the dataset from CMA is greater than that by using the dataset from JTWC by about 5%. vA-T estimated using the autoregressive type of model and the beta-advection model indicates that their differences are up to 10% and 12% for T equals 50 and 100 years if the dataset from CMA is considered.

One-shot multi-object tracking using CNN-based networks with spatial-channel attention mechanism

Deep learning algorithms for multi-object tracking have made great progress and have powered the emergence of state-of-the-art models to address multi-object tracking problems. Though a lot of efforts have been made, false detections (named “FP”) and missed detections (named “FN”) caused by inaccurate tracking still cannot be well addressed especially in extremely crowded driving situations. To address these problems, we develop a new online one-shot multi-object tracking system based on convolutional neural networks with spatial-channel attention mechanism. Firstly, we propose a feature combination module (FCM) that uses dilated convolution to obtain different receptive fields to adapt to the deformation of the targets, instead of introducing a large number of parameters to deal with the problem of target scale transformation like the recent feature pyramid network. Then, an attention mechanism network (AM-Net) is designed to allow the model to dynamically focus on certain parts of the input that help perform the task and ignore irrelevant information. Finally, we introduce a combination of triple loss and online instance matching loss (TOIM Loss) to distinguish similar instances within a class. Our proposed method is evaluated in three commonly used multi-object tracking datasets including 2DMOT15, MOT16 and MOT20. The results show that our proposed method is generally superior to the compared models.

Dynamic feature fusion with spatial-temporal context for robust object tracking

Feature fusion has been widely used for improving the tracking performance. However, how to effectively analyze the characteristics of different visual features to realize dynamical feature fusion is still a challenging task. In this paper, we propose a spatial-temporal context-based dynamic feature fusion method (STCDFF) with the correlation filters framework for object tracking. The proposed STCDFF method exploits spatial-temporal context to deeply analyze the characteristics of multiple visual features (e.g., HOG, Color-Names and CNN features) to perform feature fusion. On the one hand, spatial context is employed to evaluate the discriminative ability of different features to distinguish the target object from the background. On the other hand, temporal context is utilized to consider the representative ability of different features to capture significant appearance changes of the target object. The weight of a feature is decided by both its discriminative ability and representative ability. By exploring spatial-temporal context for feature fusion, the STCDFF method can fully utilize the strengths of different features to handle complex appearance changes and background clutters to achieve better performance. Extensive experiments on multiple object tracking datasets prove that our STCDFF method performs competitively against several popular tracking methods.

Travel routes to remote ocean targets reveal the map sense resolution for a marine migrant.

How animals navigate across the ocean to isolated targets remains perplexing greater than 150 years since this question was considered by Charles Darwin. To help solve this long-standing enigma, we considered the likely resolution of any map sense used in migration, based on the navigational performance across different scales (tens to thousands of kilometres). We assessed navigational performance using a unique high-resolution Fastloc-GPS tracking dataset for post-breeding hawksbill turtles (Eretmochelys imbricata) migrating relatively short distances to remote, isolated targets on submerged banks in the Indian Ocean. Individuals often followed circuitous paths (mean straightness index = 0.54, range 0.14–0.93, s.d. = 0.23, n = 22), when migrating short distances (mean beeline distance to target = 106 km, range 68.7–178.2 km). For example, one turtle travelled 1306.2 km when the beeline distance to the target was only 176.4 km. When off the beeline to their target, turtles sometimes corrected their course both in the open ocean and when encountering shallow water. Our results provide compelling evidence that hawksbill turtles only have a relatively crude map sense in the open ocean. The existence of widespread foraging and breeding areas on isolated oceanic sites points to target searching in the final stages of migration being common in sea turtles.

Space‐use patterns of green turtles in industrial coastal foraging habitat: Challenges and opportunities for informing management with a large satellite tracking dataset

Abstract Increasing overlap between anthropogenic activities and wildlife can lead to problematic human–wildlife interactions. To manage these, an understanding of animal space‐use patterns, with sufficient temporal and spatial detail is required. Satellite telemetry can provide such detailed data; however, the cost of tracking units places a significant limitation on sample size. Satellite tracks for 72 green turtles were consolidated through collaboration with multiple entities over 8 years at seven sites within a large industrial port contributing to an ecological monitoring initiative to minimize impacts of planned developments. This study aims to determine the minimum number of satellite‐tracked green turtles required to represent spatial distribution patterns in the foraging ground and to evaluate factors underpinning differences in distribution and site fidelity metrics to inform appropriate management strategies. An autocorrelated kernel density estimator was used to construct 95% utilization distributions for individual turtles during each calendar season. Percentage overlap between pairs of seasonal utilization distributions was calculated as a measure of short‐term site fidelity. Mechanistic range shift analysis was applied to detect significant deviations from range residency behaviour. Green turtles exhibited spatially confined ranges and remained faithful to their foraging area for periods of up to 260 days. Range size was significantly different between microhabitats and study years. Only 16 individuals (22% of tracked turtles) performed significant range shifts, indicating that occupied areas represent important habitats, and most turtles are unlikely to adjust their space‐use in response to anthropogenic or natural disturbances. Although this dataset represents an atypically large sample of satellite tracked individuals, representative data were obtained at only two key sites. This study highlights the importance of evaluating clear objectives when sampling animals for satellite telemetry studies to obtain representation of sites, periods of interest, or age and sex cohorts.

Modeling Urban Mode Choice Behavior During the COVID-19 Pandemic in Switzerland Using Mixed Multiple Discrete-Continuous Extreme Value Models

This paper describes and models the behavioral response to the COVID-19 pandemic in Switzerland. The MOBIS-COVID GPS tracking dataset, which includes a pre-pandemic reference base, is used. Trip-level data are transformed in weekly distance proportions per mode per week, and the data are modeled using a mixed multiple discrete-continuous extreme value (MMDCEV) model. Four distinct segments are derived, from September 2019 until the end of 2020, and used to uncover natural and forced behavioral adaptations. The descriptive and model estimation results confirm the trends partly observed around the globe, that is, a large decrease in public transport usage, recovered car usage, and a cycling boom. Behavioral insights are further provided as well as policy recommendations.

Why VR Games Sickness? An Empirical Study of Capturing and Analyzing VR Games Head Movement Dataset

Virtual reality (VR) technology is gaining popularity in a variety of fields, including education, games, movies, medicine, and engineering, to name a few. 360$^\circ$∘ VR video could provide an immersive experience and attract more researchers’ and developers’ attention. Some literature focused on the head movement when users watched 360$^\circ$∘ videos and released head tracking datasets. With the popularity of VR games, how the game contexts influence players’ head movement and the effect of head movement on VR sickness is a topic worth studying. In this article, we collected a head movement dataset of 30 participants while playing five different VR games (Aircar, Beat Saber, Moss, Arizona Sunshine, and SUPERHOT), and the participants filled the Simulator Sickness Questionnaire (SSQ) after playing VR games. Finally, we analyzed the SSQ scores and the impact of VR games on VR sickness. To the best of our knowledge, this is the first available head movement trajectory dataset based on playing several types of VR games.

Multibranch Adaptive Fusion Network for RGBT Tracking

RGBT tracking has been increasingly investigated in visual tracking due to the strong complementary nature of visible and infrared images. However, in the established RGBT tracking algorithms, multiscale information has not been well exploited and utilized, which limits the performance of the tracker. In this paper, a novel multibranch adaptive fusion network is proposed, which aggregates multiscale information from multiple branches. Specifically, our backbone network draws on the modified VGG-M. To extract the multiscale features, we design a multiscale adapter, which adds two small convolution kernel branches to the backbone in each layer and each modality in a parallel manner. We also design a multibranch fusion module to adaptively aggregate the features from multiple branches and the previous layer. Moreover, we propose a multimodal fusion module for aggregating features between modalities, which could mitigate the impact of noise from low-quality sources. Finally, many results on two recent RGBT tracking datasets show that our method significantly outperforms other state-of-the-art tracking methods.

Brain-inspired multiple-target tracking using Dynamic Neural Fields

Despite considerable progress in the field of automatic multi-target tracking, several problems such as data association remained challenging. On the other hand, cognitive studies have reported that humans can robustly track several objects simultaneously. Such circumstances happen regularly in daily life, and humans have evolved to handle the associated problems. Accordingly, using brain-inspired processing principles may contribute to significantly increase the performance of automatic systems able to follow the trajectories of multiple objects. In this paper, we propose a multiple-object tracking algorithm based on dynamic neural field theory which has been proven to provide neuro-plausible processing mechanisms for cognitive functions of the brain. We define several input neural fields responsible for representing previous location and orientation information as well as instantaneous linear and angular speed of the objects in successive video frames. Image processing techniques are applied to extract the critical object features including target location and orientation. Two prediction fields anticipate the objects’ locations and orientations in the upcoming frame after receiving excitatory and inhibitory inputs from the input fields in a feed-forward architecture. This information is used in the data association and labeling process. We tested the proposed algorithm on a zebrafish larvae segmentation and tracking dataset and an ant-tracking dataset containing non-rigid objects with spiky movements and frequently occurring occlusions. The results showed a significant improvement in tracking metrics compared to state-of-the-art algorithms.

A dataset of eye gaze images for calibration-free eye tracking augmented reality headset

Eye tracking is a widely used technique. To enhance eye gaze estimation in different contexts, many eye tracking datasets have been proposed. However, these datasets depend on calibrations in data capture and its applications. We seek to construct a dataset that enables the design of a calibration-free eye tracking device irrespective of users and scenes. To reach this goal, we present ARGaze, a dataset with 1,321,968 pairs of eye gaze images at 32 × 32 pixel resolution and 50 corresponding videos of world views based on a replicable augmented reality headset. The dataset was captured from 25 participants who completed eye gaze tasks for 30 min in both real-world and augmented reality scenes. To validate the dataset, we compared it against state-of-the-art eye gaze datasets in terms of effectiveness and accuracy and report that the ARGaze dataset achieved record low gaze estimation error by 3.70 degrees on average and 1.56 degrees on specific participants without calibrations to the two scenes. Implications for generalising the use of the dataset are discussed.

Visual Object Tracking for Unmanned Aerial Vehicles Based on the Template-Driven Siamese Network

Visual object tracking for unmanned aerial vehicles (UAV) is widely used in many fields such as military reconnaissance, search and rescue work, film shooting, and so on. However, the performance of existing methods is still not very satisfactory due to some complex factors including viewpoint changing, background clutters and occlusion. The Siamese trackers, which offer a convenient way of formulating the visual tracking problem as a template matching process, have achieved success in recent visual tracking datasets. Unfortunately, these template match-based trackers cannot adapt well to frequent appearance change in UAV video datasets. To deal with this problem, this paper proposes a template-driven Siamese network (TDSiam), which consists of feature extraction subnetwork, feature fusion subnetwork and bounding box estimation subnetwork. Especially, a template library branch is proposed for the feature extraction subnetwork to adapt to the changeable appearance of the target. In addition, a feature aligned (FA) module is proposed as the core of feature fusion subnetwork, which can fuse information in the form of center alignment. More importantly, a method for occlusion detection is proposed to reduce the noise caused by occlusion. Experiments were conducted on two challenging benchmarks UAV123 and UAV20L, the results verified the more competitive performance of our proposed method compared to the existing algorithms.

Mapping Exposure Diversity: The Divergent Effects of Algorithmic Curation on News Consumption

Abstract Diversity is a crucial precondition for a democratic public discourse. In today’s high-choice media environments, exposure to diverse news is largely determined by individuals’ personal selection. Yet these decisions are increasingly shaped by online platforms, whose curation mechanisms may serve to expand or contract the diversity of encountered content. In a major extension of existing research, we show that positive short-term effects of platforms mask detrimental long-term effects. Drawing on a four-month tracking dataset and a comprehensive content analysis covering the online news consumption of over 10,000 German citizens, we demonstrate that even though short-term usage of platforms uniformly increases exposure diversity, long-term reliance can lead to decreases. In addition, platforms vary in their influences: News aggregators are beneficial to exposure diversity, while Twitter and search engines have a limiting effect; Facebook offers no significant influence.

Socially aware robot navigation in crowds via deep reinforcement learning with resilient reward functions

Robots navigating in a robot–human coexisting environment need to optimize their paths not only for task-related performance (e.g. safety and efficiency) but also for their social compliance to other pedestrians. This is a crucial yet challenging task. Previous work has shown the power of deep reinforcement learning (DRL) techniques by employing them to train efficient policies for robot navigation. However, their performance deteriorates when the crowd size grows. We cope with this problem by allowing the robot to keep an adapting distance from the pedestrians and perform safe navigation even in high density environments. We first derive a quantitative formula representing the relationship between uncomfortable distance and pedestrian density from a real-word tracking dataset. Then this formula is applied in reward shaping of DRL to get resilient reward functions (R2F). Qualitative and quantitative evaluation results demonstrate that our method outperforms state-of-the-art methods in both low and high pedestrian density environments.

Feature adaptation-based multipeak-redetection spatial-aware correlation filter for object tracking

Visual tracking has always been an important research topic in the field of computer vision. During tracking, cluttered backgrounds, deformations, and occasional occlusion inevitably cause unpredictable appearance changes in the target, making the task very challenging. Traditional methods often tackle this problem using fixed central region constraints, simple combinations of features and independent classifiers. However, existing constraints are not very accurate; the simple mechanism of combining features lacks descriptive robustness, and the independent classifiers require complex additional training. Therefore, in this paper, we propose a correlation filter tracker called the feature adaptation-based multipeak-redetection spatial-aware correlation filter (FMCF) to accurately track targets that often change in appearance. In this tracker, the correlation filter has a dynamic spatial constraint that effectively reduces the response of the filter to the cluttered background area. To address target deformation, a new feature adaptation-based training method is designed, and the final response is a combination of the correlation filter and an independent statistical color model. Finally, a multipeak-redetection scheme, which can find the missing targets within multiple peaks of the final response map, is proposed to handle occlusion. In particular, the statistical color model is employed in the spatial-aware filter construction, the final response calculation, and the multipeak-redetection procedure to improve the robustness of the tracker, because of its ability to generate pixel-level color probabilities. Extensive experimental results on public tracking datasets prove that our proposed tracker can perform better than state-of-the-art trackers. Moreover, compared to the baseline tracker, the precision rate and success rate of FMCF are increased by 6.5% and 4.8% on the OTB100 benchmark and by 12.7% and 8.2% on the TC128 benchmark, respectively.

Long‐time target tracking algorithm based on re‐detection multi‐feature fusion

This paper considers the problem of long-term target tracking in complex scenes when tracking failures are unavoidable due to illumination change, target deformation, scale change, motion blur, and other factors. More specifically, a target tracking algorithm, called re-detection multi-feature fusion, is proposed based on the fusion of scale-adaptive kernel correlation filtering and re-detection. The target tracking algorithm trains three kernel correlation filters based on the histogram of oriented gradients, colour name, and local binary pattern features and then obtains the fusion weight of response graphs corresponding to different features based on average peak correlation energy criterion and uses weighted average to complete the position estimation of the tracked target. In order to deal with the problem that the target is occluded and disappears in the tracking process, a random fern classifier is trained to perform re-detection when the target is occluded. After comparing the OTB-50 target tracking dataset, the experimental results show that the proposed tracker can track the target well in the occlusion attribute video sequence in the OTB-100 test dataset and has a certain improvement in tracking accuracy and success rate compared with the traditional correlation filter tracker.

MMSTN: A Multi‐Modal Spatial‐Temporal Network for Tropical Cyclone Short‐Term Prediction

AbstractForecasting the trajectory and intensity of tropical cyclones (TCs) is important in disaster mitigation as TC usually causes huge damages. However, it remains a substantial challenge due to the limited understanding of TC complexity. Still, TCs have been observed and recorded for several decades, and so can be predicted if viewed as a spatial‐temporal prediction problem with a huge amount of existing data. We propose a novel TC trajectory and intensity short‐term prediction method: Multi‐Modal Spatial‐temporal Networks (MMSTN). It not only predicts the TC's central pressure, winds, and the location of its center, but also forecasts the TC's varied possible tendencies. Experiments were conducted on the China Meteorological Administration Tropical Cyclone Best Track Dataset. Experimental results show that the proposed MMSTN outperformed state‐of‐the‐art methods as well as the official prediction method of the China Central Meteorological Observatory, in intensity prediction and 6 hr trajectory prediction.

SimCLRT: A Simple Framework for Contrastive Learning of Rumor Tracking

As the second stage of the rumor defeat task pipeline, rumor tracking aims to filter tweets related to a specific event. However, due to the different levels of attention of different events, events related to fewer tweets may be masked by related to more tweets. In some research, the researchers give up detecting the events containing a small number of tweets for better results. To this end, we propose a Simple Framework for Contrastive Learning of Rumor Tracking (SimCLRT)-a novel rumor tracking framework that uses contrastive learning to alleviate the cover between tweets. SimCLRT contains three variants SimCLRT-CNN, SimCLRT-Linear, and SimCLRT-RNN. We conduct experiments on the two commonly used rumor tracking datasets PHEME and RumorEval. The results show that SimCLRT completely defeated baselines. Like the detection performance on the events that contain many tweets, SimCLRT can also effectively detect events containing a small number of tweets. Furthermore, we compare and analyze the performance of SimCLRT variants. SimCLRT-CNN is the model that performs best in our experiments. Although SimCLRT-Linear has a slight advantage on the RumorEval dataset, its robustness is weaker than SimCLRT-RNN and SimCLRT-CNN. If in a long text environment, we consider SimCLRT-RNN will perform more competitively.