Localization Task Research Articles

FeMLoc: Federated Meta-Learning for Adaptive Wireless Indoor Localization Tasks in IoT Networks

FeMLoc: Federated Meta-Learning for Adaptive Wireless Indoor Localization Tasks in IoT Networks

Technical Note: Neural Network Architectures for Self-Supervised Body Part Regression Models with Automated Localized Segmentation Application.

The advancement of medical image deep learning necessitates tools that can accurately identify body regions from whole-body scans to serve as an essential pre-processing step for downstream tasks. Typically, these deep learning models rely on labeled data and supervised learning, which is labor-intensive. However, the emergence of self-supervised learning is revolutionizing the field by eliminating the need for labels. The purpose of this study was to compare neural network architectures of self-supervised models that produced a body part regression (BPR) slice score to aid in the development of anatomically localized segmentation models. VGG, ResNet, DenseNet, ConvNext, and EfficientNet BPR models were implemented in the MONAI/Pytorch framework. Landmark organs were correlated to slice scores and mean absolute error (MAE) was calculated from the predicted slice and the actual slice of various organ landmarks. Four localized DynUNet segmentation models (thorax, upper abdomen, lower abdomen, and pelvis) were developed using the BPR slice scores. Dice similarity coefficient (DSC) was compared between the localized and baseline segmentation models. The best performing BPR model was the EfficientNet architecture with an overall 3.18 MAE, compared to the VGG baseline model with a MAE of 6.29. The localized segmentation model significantly outperformed the baseline in 16 out of 20 organs with a DSC of 0.88. Enhanced neural networks like EfficientNet have a large performance increase in localizing anatomical structures in a CT compared in BPR task. Utilizing BPR slice score is shown to be effective in anatomically localized segmentation tasks with improved performance.

Audio-visual event localization with dual temporal-aware scene understanding and image-text knowledge bridging

Audio-visual event localization (AVEL) task aims to judge and classify an audible and visible event. Existing methods devote to this goal by transferring pre-trained knowledge as well as understanding temporal dependencies and cross-modal correlations of the audio-visual scene. However, most works comprehend the audio-visual scene from an entangled temporal-aware perspective, ignoring the learning of temporal dependency and cross-modal correlation in both forward and backward temporal-aware views. Recently, transferring the pre-trained knowledge from Contrastive Language-Image Pre-training model (CLIP) has shown remarkable results across various tasks. Nevertheless, since audio-visual knowledge of the AVEL task and image-text alignment knowledge of the CLIP exist heterogeneous gap, how to transfer the image-text alignment knowledge of CLIP into AVEL field has barely been investigated. To address these challenges, a novel Dual Temporal-aware scene understanding and image-text Knowledge Bridging (DTKB) model is proposed in this paper. DTKB consists of forward and backward temporal-aware scene understanding streams, in which temporal dependencies and cross-modal correlations are explicitly captured from dual temporal-aware perspectives. Consequently, DTKB can achieve fine-grained scene understanding for event localization. Additionally, a knowledge bridging (KB) module is proposed to simultaneously transfer the image-text representation and alignment knowledge of CLIP to AVEL task. This module regulates the ratio between audio-visual fusion features and CLIP’s visual features, thereby bridging the image-text alignment knowledge of CLIP and the audio-visual new knowledge for event category prediction. Besides, the KB module is compatible with previous models. Extensive experimental results demonstrate that DTKB significantly outperforms the state-of-the-arts models.

Differentiating Postural and Kinetic Tremor Responses to Deep Brain Stimulation in Essential Tremor.

While deep brain stimulation (DBS) targeting the ventral intermediate nucleus (VIM) of thalamus or posterior subthalamic area (PSA) can suppress forms of action tremor in people with Essential Tremor, previous studies have suggested postural tremor may respond more robustly than kinetic tremor to DBS. In this study, we aimed to more precisely quantify the (1) onset/offset dynamics and (2) steady-state effects of VIM/PSA-DBS on postural and kinetic tremor. Tremor data from wireless inertial measurement units were collected from 11 participants with ET (20 unilaterally assessed DBS leads). Three postural hold tasks and one kinetic task were performed with stimulation turned off, in 2-min intervals after enabling unilateral DBS at the clinician-optimized DBS setting (15 min), and in 2-min intervals following cessation of DBS (5 min). At baseline, kinetic tremor had significantly higher amplitudes, standard deviation, and frequency than postural tremor (P < 0.001). DBS had a more robust acute effect on postural tremors (54% decrease, P < 0.001), with near immediate tremor suppression in amplitude and standard deviation, but had non-significant improvement of kinetic tremor on the population-level across the wash-in period (34% decrease). Tremor response was not equivalent between wash-in and wash-out timepoints and involved substantial individual variability including task-specific rebound or long wash-out effects. Programming strategies for VIM/PSA-DBS should consider the individual temporal and effect size variability in postural versus kinetic tremor improvement. Improved targeting and programming strategies around VIM and PSA may be necessary to equivalently suppress both postural and kinetic tremors.

Artificial intelligence as planetary assemblages of coloniality: The new power architecture driving a tiered global data economy

We present a framework for viewing artificial intelligence (AI) as planetary assemblages of coloniality that reproduce dependencies in how it co-constitutes and structures a tiered global data economy. We use assemblage thinking to map the coloniality of power to demonstrate how AI stratifies across knowledge, geographies, and bodies to influence development and economic trajectories, impact workers, reframe domestic industrial policies, and reconfigure the international political economy. Our post-colonial framework unpacks AI through its (1) global, (2) meso, and (3) local layers, and further dissects how these layers are vertically integrated, each with its horizontal dependencies. At (1) the global layer of international political economy maps a new digital bipolarity expressing Sino and American global digital corporations’ strategic and dominant positions in shaping a tiered global data economy. Then, at (2) the meso layer, we have a mosaic of domestic industrial policies that fund, frame markets, and develop AI talent across industries, sectors, and organizations to competitively integrate into AI value chains. Finally, incorporating into these are (3) the localized labor processes and tasks, where workers and users enact various AI-mediated tasks and practices driving further value extraction. We traced how AI is an interlaced system of power that reshapes knowledge, geographies, and bodies into dependencies that reinforce stratifications in developing underdevelopment. This commentary maps the current digital realities by laying out an uneven techno-geoeconomic power architecture driving a tiered global data economy and opening new research avenues to examine AI as planetary assemblages of coloniality.

Applications of cluster-based transfer learning in image and localization tasks

Transfer learning can address the issue of insufficient labels in machine learning. Using knowledge in a labeled domain (source domain) can assist in acquiring and learning knowledge in a domain (target domain) that lacks some or all labels. In this paper, we propose a new cluster-based semi-supervised transfer learning (CBSSTL) under a new assumption that samples in the target domain are unlabeled but contain cluster information. Furthermore, we propose a new transfer learning framework and a method for fine-tuning parameters. We tested and compared the proposed method with other unsupervised and semi-supervised transfer learning methods on well-known image datasets. The experimental results demonstrate the effectiveness of the proposed method. Additionally, we created a localization dataset for transfer learning. Finally, we tested and analyzed the proposed method on this dataset. Its particularly challenging nature makes it difficult for our method to work effectively.

IDL-LTSOJ: Research and implementation of an intelligent online judge system utilizing DNN for defect localization

The evolution of artificial intelligence has thrust the Online Judge (OJ) systems into the forefront of research, particularly within programming education, with a focus on enhancing performance and efficiency. Addressing the shortcomings of the current OJ systems in coarse defect localization granularity and heavy task scheduling architecture, this paper introduces an innovative Integrated Intelligent Defect Localization and Lightweight Task Scheduling Online Judge (IDL-LTSOJ) system. Firstly, to achieve token-level fine-grained defect localization, a Deep Fine-Grained Defect Localization (Deep-FGDL) deep neural network model is developed. By integrating Bidirectional Long Short-Term Memory (BiLSTM) and Bidirectional Gated Recurrent Unit (BiGRU), this model extracts fine-grained information from the abstract syntax tree (AST) of code, enabling more accurate defect localization. Subsequently, we propose a lightweight task scheduling architecture to tackle issues, such as limited concurrency in task evaluation and high equipment costs. This architecture integrates a Kafka messaging system with an optimized task distribution strategy to enable concurrent execution of evaluation tasks, substantially enhancing system evaluation efficiency. The experimental results demonstrate that the Deep-FGDL model improves the accuracy by 35.9% in the Top-20 rank compared to traditional machine learning benchmark methods for fine-grained defect localization tasks. Moreover, the lightweight task scheduling strategy notably reduces response time by nearly 6000ms when handling 120 task volumes, which represents a significant improvement in evaluation efficiency over centralized evaluation methods.

Online indoor visual odometry with semantic assistance under implicit epipolar constraints

Among solutions to the tasks of indoor localization and reconstruction, compared with traditional SLAM (Simultaneous Localization And Mapping), learning-based VO (Visual Odometry) has gained more and more popularity due to its robustness and low cost. However, the performance of existing indoor deep VOs is still limited in comparison with their outdoor counterparts mainly owing to large areas of textureless regions and complex indoor motions containing much more rotations. In this paper, the above two challenges are carefully tackled with the proposed SEOVO (Semantic Epipolar-constrained Online VO). On the one hand, as far as we know, SEOVO is the first semantic-aided VO under an online adaptive framework, which adaptively reconstructs low-texture planes without any supervision. On the other hand, we introduce the epipolar geometric constraint in an implicit way for improving the accuracy of pose estimation without destroying the global scale consistency. The efficiency and efficacy of SEOVO have been corroborated by extensive experiments conducted on both public datasets and our collected video sequences.

The Comparison of Two Multitasking Approaches to Cognitive Training in Patients after Coronary Bypass Surgery: Theta Activity Changes and &lt;i&gt;sLORETA&lt;/i&gt; Analysis Data

The study investigated the changes in theta activity and localization of its sources by standardized low resolution brain electromagnetic tomography (sLORETA) in patients who have underwent two variants of multitasking cognitive training (CT) in the early postoperative period of coronary artery bypass grafting (CABG). Two groups were formed in a pseudo-random way, which differed according to the type of motor problem used: CT I (n = 27) – a postural balance task and CT II (n = 27) – a simple visual-motor reaction. Cognitive tasks were the same for both groups (counting backwards, verbal fluency, and unusual uses for common objects). Daily sessions of CT were held from the 3rd to 4th day after CABG, with a duration of 5 minutes on the 1st day of training and up to 20 minutes on the 6th to 7th day of training. The current density of theta rhythm sources was lower before CABG than after surgery in the CT II group only. The most significant differences are in the Brodmann area 31, the parietal occipital lobes and precuneus, which may indicate damage associated with cardiac surgery. This effect was not observed in the CT I group. The results of our study demonstrated the informativeness of sLORETA indicators to determine an effective cognitive recovery option after CABG. The reduction of the severity of damaging effects of CABG during training using cognitive tasks and postural balance task was shown. Further research is needed to determine the optimal mode and duration of cognitive training to maximize the functional reserves of such patients.

Head rotations follow those of a truncated Fick gimbal during an auditory guided visual search task.

Recent interest in dynamic sound localisation models has created a need to better understand the head movements made by humans. Previous studies have shown that static head positions and small oscillations of the head obey Donders' law: for each facing direction there is one unique three-dimensional orientation. It is unclear whether this same constraint applies to audiovisual localisation, where head movement is unrestricted and subjects may rotate their heads depending on the available auditory information. In an auditory guided visual search task, human subjects were instructed to localise an audiovisual target within a field of visual distractors in the frontal hemisphere. During this task, head and torso movements were monitored using a motion capture system. Head rotations were found to follow Donders' law during search tasks. Individual differences were present in the amount of roll that subjects deployed, though there was no statistically significant improvement in model performance when including these individual differences in a gimbal model. The roll component of head rotation could therefore be predicted with a truncated Fick gimbal, which consists of a pitch axis nested within a yaw axis. This led to a reduction from three to two degrees of freedom when modelling head movement during localisation tasks.

Characterization of implicit and explicit mind-reading in children with autism based on eye movements.

This study aimed to investigate differences in mind-reading abilities between children with autism and typically developing children across various tasks. Sixteen children with autism (aged 5-8 years) were compared to 16 typically developing children matched in language ability. The unexpected location task and unexpected content task were used to assess implicit and explicit mind-reading abilities using an eye tracker and illustrated storybooks. For implicit mind, using differential looking scores shows a no significant difference between the scores of children with autism and typically developing children in the implicit tasks (P=0.399). However, the pupil size show some significant difference between two groups. Second, for the explicit, a significant difference between the scores of children with autism and typically developing children in the explicit tasks (P=0.006). Additionally, only a significant correlation between implicit and explicit mind-reading abilities in children with autism in the unexpected location task was found. The mechanism of implicit mind-reading still not very clear. Pupil-Linked arousal response can be consider as a further tool. Further research on implicit and explicit mind-reading abilities is warranted.

Vibration Thresholds Using Conventional Audiometry are Clinically Useful Indicators of Postural Instability in Older Adults.

Falls are a major health concern with potentially dramatic consequences for people over 65 years of age. One crucial determinant in the risk of falls in older adults is postural control, a complex process that requires the contribution of different sensory modalities, namely visual, vestibular, auditory, and somatosensory. While there are well-established methods to screen for age-related vision, hearing, tactile, and vestibular impairments, there are very few widely available methods to screen for somatosensory function, but studies indicate that ankle audiometry (vibration thresholds) using a common B-71 bone vibrator can serve that purpose. To date, unfortunately, this technique has received little attention as a tool to measure postural instability in older adults. The objective of the present study was to examine postural control in older adults with and without degradation of the somatosensory functions, as determined with ankle audiometry. This was standard group comparison. In total, 36 healthy elderly aged between 65 and 80 years old were divided into two groups (low vibration threshold [n = 18] and high vibration threshold [n = 18]). Standard audiometry, video head impulse test, vibration thresholds (big toe, ankle, and tibia), and static postural control task using a force platform were performed. Greater postural instability in participants with higher (worse) vibration thresholds as compared with participants with lower (better) vibration thresholds was observed even though both groups were comparable on hearing threshold and vestibular function. The results indicate that performing a simple vibration threshold evaluation, using a clinically available B-71 with a cut-off value of 42 dB hearing loss, could be an effective, fast, and easy-to-use procedure for detecting people at risk of falls.

Cephalometric landmark annotation using transfer learning: Detectron2 and YOLOv8 baselines on a diverse cephalometric image dataset

BackgroundCephalometric landmark annotation is a key challenge in radiographic analysis, requiring automation due to its time-consuming process and inherent subjectivity. This study investigates the application of advanced transfer learning techniques to enhance the accuracy of anatomical landmarks in cephalometric images, which is a vital aspect of orthodontic diagnosis and treatment planning. MethodsWe assess the suitability of transfer learning methods by employing state-of-the-art pose estimation models. The first framework is Detectron2, with two baselines featuring different ResNet backbone architectures: rcnn_R_50_FPN_3x and rcnn_R_101_FPN_3x. The second framework is YOLOv8, with three variants reflecting different network sizes: YOLOv8s-pose, YOLOv8m-pose, and YOLOv8l-pose. These pose estimation models are adopted for the landmark annotation task. The models are trained and evaluated on the DiverseCEPH19 dataset, comprising 1692 radiographic images with 19 landmarks, and their performance is analyzed across various images categories within the dataset. Additionally, the study is extended to a benchmark dataset of 400 images to investigate how dataset size impacts the performance of these frameworks. ResultsDespite variations in objectives and evaluation metrics between pose estimation and landmark localization tasks, the results are promising. Detectron2's variant outperforms others with an accuracy of 85.89%, compared to 72.92% achieved by YOLOv8's variant on the DiverseCEPH19 dataset. This superior performance is also observed in the smaller benchmark dataset, where Detectron2 consistently maintains higher accuracy than YOLOv8. ConclusionThe noted enhancements in annotation precision suggest the suitability of Detectron2 for deployment in applications that require high precision while taking into account factors such as model size, inference time, and resource utilization, the evidence favors YOLOv8 baselines.

Reinforcement learning-based method for type B aortic dissection localization

In the segmentation of aortic dissection, there are issues such as low contrast between the aortic dissection and surrounding organs and vessels, significant differences in dissection morphology, and high background noise. To address these issues, this paper proposed a reinforcement learning-based method for type B aortic dissection localization. With the assistance of a two-stage segmentation model, the deep reinforcement learning was utilized to perform the first-stage aortic dissection localization task, ensuring the integrity of the localization target. In the second stage, the coarse segmentation results from the first stage were used as input to obtain refined segmentation results. To improve the recall rate of the first-stage segmentation results and include the segmentation target more completely in the localization results, this paper designed a reinforcement learning reward function based on the direction of recall changes. Additionally, the localization window was separated from the field of view window to reduce the occurrence of segmentation target loss. Unet, TransUnet, SwinUnet, and MT-Unet were selected as benchmark segmentation models. Through experiments, it was verified that the majority of the metrics in the two-stage segmentation process of this paper performed better than the benchmark results. Specifically, the Dice index improved by 1.34%, 0.89%, 27.66%, and 7.37% for each respective model. In conclusion, by incorporating the type B aortic dissection localization method proposed in this paper into the segmentation process, the overall segmentation accuracy is improved compared to the benchmark models. The improvement is particularly significant for models with poorer segmentation performance.

Optimization of artificial intelligence in localized big data real-time query processing task scheduling algorithm

Optimization of artificial intelligence in localized big data real-time query processing task scheduling algorithm

Real-time spatiotemporal action localization algorithm using improved CNNs architecture

This paper aims to propose a faster and more accurate network for human spatiotemporal action localization tasks. Like the YOWO model, we also use convolutional neural networks (CNNs) for feature extraction, but our model differs from YOWO in three significant ways: firstly, we don’t use the feature fusion strategy, we only use spatial features extracted by 2D CNNs for action localization and spatiotemporal features extracted by 3D CNNs for action recognition; secondly, we make an improvement to the 2D CNNs network by introducing a coordinate attention mechanism and utilize the CIoU loss instead of the coordinate offset loss for bounding box regression; thirdly, we provide a more lightweight and faster spatiotemporal action localization architecture, which reduces the number of parameters by 21.76 million and achieves a speed of 39 fps on 16-frame input clips compared to the YOWO model. We test our model’s performance on three public datasets: UCF-Sports, JHMDB-21 and UCF101-24. Compared with the YOWO model, we improve frame-mAP (@IoU 0.5) by 17.09% and 7.15% on the UCF-Sports and JHMDB-21 datasets, and for video-mAP, we improve by 2.7%, 8.7% and 14.4% at IoU thresholds of 0.2, 0.5 and 0.75 on the JHMDB-21 dataset.

Cognitive functions explain discrete parameters of normal walking and dual-task walking, but not postural sway in quiet stance among physically active older people.

Postural control is dependent on the central nervous system's accurate interpretation of sensory information to formulate and execute adequate motor actions. Research has shown that cognitive functions are associated with both postural control and fall risk, but specific associations are not established. The aim of this study was to explore how specific components of everyday postural control tasks are associated with both general and specific cognitive functions. Forty-six community-dwelling older adults reported their age, sex, physical activity level, falls and fall-related concerns. The following cognitive aspects were assessed: global cognition, executive functions, processing speed and intraindividual variability. Postural control was quantified by measuring postural sway in quiet stance, walking at a self-selected pace, and walking while performing a concurrent arithmetical task. Separate orthogonal projections of latent structures models were generated for each postural control outcome using descriptive and cognitive variables as explanatory variables. Longer step length and faster gait speed were related to faster processing speed and less intraindividual variability in the choice reaction test. Moreover, longer step length was also related to less fall-related concerns and less severe fall-related injuries, while faster gait speed was also related to female sex and poorer global cognition. Lower dual-task cost for gait speed was explained by the executive function inhibition and faster processing speed. Postural sway in quiet stance was not explained by cognitive functions. Cognitive functions explained gait speed and step length during normal walking, as well as the decrease of gait speed while performing a concurrent cognitive task. The results suggest that different cognitive processes are important for different postural control aspects. Postural sway in quiet stance, step time and gait variability seem to depend more on physical and automatic processes rather than higher cognitive functions among physically active older people. The relationships between cognitive functions and postural control likely vary depending on the specific tasks and the characteristics of different populations.

Optimized Factor Graph for Tightly-Coupled LiDAR/IMU Localization in Underground Parking Garages

Abstract. To address the challenge of intelligent vehicle localization in underground parking structures due to the loss of GNSS signals, this paper introduces a method to address this issue by developing a novel localization framework known as GF-LIO, which denotes a tightly-coupled fusion of LiDAR and IMU data, innovatively combining the Interactive Extended State Kalman Filter (IESKF) with a factor graph to enhance the localization process and solve the problem of GNSS signal loss in underground parking lots. The GF-LIO model commences with a strategic feature selection process, facilitated by a greedy algorithm that prioritizes environmental cues within the point cloud data. This method effectively filters out redundant features, thereby enhancing the saliency of retained features and subsequently improving the robustness of the localization process. Following feature selection, the model integrates LiDAR and IMU measurements utilizing the IESKF algorithm, ensuring a cohesive fusion of sensor data and bolstering attitude estimation accuracy. The culmination of the GF-LIO framework involves factor graph optimization, a sophisticated technique that synthesizes LiDAR odometry, IMU pre-integration factors, and loop closure detection factors. This optimization step enhances the overall precision and consistency of the localization process, resulting in superior performance compared to existing methodologies. Experimental evaluations conducted within underground parking environments corroborate the efficacy of the GF-LIO model. Comparative analyses against established approaches such as A-LOAM, LeGO-LOAM, LIO-LOAM, and FAST-LIO demonstrate a notable performance improvement exceeding 12.53%. The proposed model adeptly integrates domain-specific environmental characteristics with multi-sensor data, thereby facilitating precise localization and map construction tasks for intelligent vehicle navigating within the intricate confines of subterranean parking structures.

Context-aware feature reconstruction for class-incremental anomaly detection and localization

With the development of deep learning, the unsupervised visual anomaly detection and localization task has gained significant attention in both academia and industry, where only normal data are used for training. Existing methods for this task typically train using all training data simultaneously. However, in practical industrial scenarios, new product classes are usually introduced incrementally, leading to the sequential availability of training data. Such scenarios demand methods for class-incremental anomaly detection and localization (CADL). The main challenge of class-incremental learning is to retain knowledge of old classes when learning new classes. In this article, we aim to effectively leverage limited exemplars of old classes to retain knowledge for the CADL task. Achieving this goal requires a model that can efficiently capture anomaly-identification-related knowledge from limited exemplars. Considering that pixel-level anomaly identification requires an understanding of the surrounding context, we treat context within inputs as valuable anomaly-identification-related knowledge and design a context-aware feature reconstruction (CFR) model to capture such knowledge. Moreover, to avoid inter-class context conflict that may arise with class increments, we design an intermediate feature organization strategy. This strategy and output-level knowledge distillation jointly form dual constraints to regularize the model at both mid-feature and output levels. Utilizing the CFR model with Dual Constraints, the proposed CFRDC can effectively retain old-class knowledge while learning new classes, thus addressing the CADL task. Experimental results on the commonly-used MVTec-AD dataset demonstrate the effectiveness and outstanding performance of the proposed method in CADL.

PGMF-VINS: Perpendicular-Based 3D Gaussian–Uniform Mixture Filter

Visual–Inertial SLAM (VI-SLAM) has a wide range of applications spanning robotics, autonomous driving, AR, and VR due to its low-cost and high-precision characteristics. VI-SLAM is divided into localization and mapping tasks. However, researchers focus more on the localization task while the robustness of the mapping task is often ignored. To address this, we propose a map-point convergence strategy which explicitly estimates the position, the uncertainty, and the stability of the map point (SoM). As a result, the proposed method can effectively improve the quality of the whole map while ensuring state-of-the-art localization accuracy. The convergence strategy mainly consists of a perpendicular-based triangulation and 3D Gaussian–uniform mixture filter, and we name it PGMF, perpendicular-based 3D Gaussian–uniform mixture filter. The algorithm is extensively tested on open-source datasets, which shows the RVM (Ratio of Valid Map points) of our algorithm exhibits an average increase of 0.1471 compared to VINS-mono, with a variance reduction of 68.8%.