Sensory Temporal Action Detection (STAD) aims to localize and classify human actions within long, untrimmed sequences captured by non-visual sensors such as WiFi or inertial measurement units (IMUs). Unlike video-based TAD, STAD poses unique challenges due to the low-dimensional, noisy, and heterogeneous nature of sensory data, as well as the real-time and resource constraints on edge devices. While recent STAD models have improved detection performance, their high computational cost hampers practical deployment. In this paper, we propose SlimSTAD, a simple yet effective framework that achieves both high accuracy and low latency for STAD. SlimSTAD features a novel Decoupled Channel Modeling (DCM) encoder, which preserves modality-specific temporal features and enables efficient inter-channel aggregation via lightweight graph attention. An anchor-free cascade predictor then refines action boundaries and class predictions in a two-stage design without dense proposals. Experiments on two real-world datasets demonstrate that SlimSTAD outperforms strong video-derived and sensory baselines by an average of 2.1 mAP, while significantly reducing GFLOPs, parameters, and latency, validating its effectiveness for real-world, edge-aware STAD deployment.

ABSTRACT Dense point cloud registration is a critical problem in computer vision and 3D reconstruction, with widespread applications in scenarios such as robotic navigation, autonomous driving, and 3D measurement. However, dense point cloud registration faces significant challenges, including high computational complexity and prolonged processing times. To address these issues, this paper proposes a point cloud registration method based on multiple neighborhood feature difference (MNFD) that employs a coarse‐to‐fine strategy to effectively enhance both registration efficiency and accuracy. The proposed method consists of two stages: coarse registration and fine registration. In the coarse registration stage, a novel feature point extraction approach based on MNFD is introduced, capable of identifying highly stable and distinctive feature points in the point cloud. These feature points are then utilized in combination with the fast point feature histogram (FPFH) algorithm to achieve an initial alignment between the target and template point clouds. In the fine registration stage, the results from the coarse alignment are refined using algorithms such as iterative closest point (ICP) to ensure both efficiency and precision during the registration process. Experiments conducted on publicly available datasets demonstrate the superiority of the proposed method compared to existing approaches.

The extraction of regions of interest (ROI) plays a vital role in enhancing the precision of target analysis and identification, especially for hyperspectral small targets in wide fields of view. While the use of deep learning for ROI extraction has demonstrated significant potential, its efficacy has been hindered by a scarcity of labeled data. This study proposes a novel ROI extraction method employing a fully convolutional network and channel-spatial attention (CSFCN) for hyperspectral small targets through self-supervised learning. First, a strategy for pseudo-label assignment is designed based on the feature similarity and spatial continuity of hyperspectral images (HSIs). Second, an unsupervised segmentation model for HSIs is established, incorporating an attention mechanism based on FCN. Finally, ROIs are extracted based on segmentation results. The experimental results on two real-world HSIs, HSIa and HSIb, show that the proposed method can effectively improve the accuracy of extracted ROIs. The overall accuracy (OA) and the intersection over union (IOU) values of the proposed CSFCN reached 25.62%, 44.44% (HSIa), and 44.76%, 100% (HSIb), far exceeding the traditional unsupervised segmentation results. The superior experimental results demonstrate that the proposed method has promising application prospects.

Despite the intense effort from the research community, state-of-the-art WiFi-based human activity recognition (HAR) performance remains unsatisfactory. Current approaches usually use individual characteristics of CSI, i.e., amplitude or phase measurements, to model the relationship between the changes of channel state information (CSI) and human activities, which lead to their failure to achieve satisfactory accuracy due to information loss. To deal with this issue, this article proposes CeHAR, a CSI-based HAR using a channel-exchanging fusion network to deep fuse the CSI amplitude and phase features to obtain the informative features for HAR. The proposed CeHAR is a parameter-free dual-characteristic fusion framework that dynamically exchanges channels between subnetworks of two kinds of characteristics to comprehensively learn informative features from both. Specifically, the proposed approach employs two subnetworks using convolutional neural networks to learn features from each characteristic of CSI. The magnitude of the batch-normalization (BN) scaling factor is used to determine the channel importance of each characteristic, and then guides the exchange process. The proposed CeHAR also shares convolutional filters, but keeps private BNs layers in different characteristics, which, as an added benefit, allows our characteristic fusion network to be nearly as compact as a single-characteristic network. Extensive real-world experiments have been conducted to evaluate the performance of our proposed CeHAR, and the experimental results illustrate that our proposed approach outperforms baselines.

Abstract Depth information plays an important role in the vision‐related activities of robots and autonomous vehicles. An effective method to obtain 3D scene information is self‐supervised monocular depth estimation, which utilizes large and diverse monocular video datasets during the training process without the need for ground‐truth data. A novel multi‐task learning strategy that uses semantic information to guide the monocular depth estimation method while maintaining self‐supervision is proposed. An improved differential direct visual odometer (DDVO) combined with Pose‐Net is applied for achieving better pose prediction. Minimum reprojection loss with auto‐masking and semantic masking is used to remove the effects of low‐texture areas and moving dynamic‐class objects within scenes. Concurrently, the semantic masking is introduced into the DDVO pose predictor to filter moving objects and reduce the matching error between monocular sequence frames. In addition, PackNet is employed as the backbone of multi‐task learning to further improve the accuracy of deep prediction. The proposed method produces state‐of‐the‐art results for monocular depth estimation on the KITTI Eigen split benchmark, even outperforming supervised methods that have been trained using ground‐truth depth.

Recently, Scene Representation Networks (SRNs) have attracted increasing attention in computer vision, due to their continuous and light-weight scene representation ability. However, SRNs generally perform poorly on low-texture image regions. Addressing this problem, we propose superpixel-guided scene representation networks in this paper, called SG-SRNs, consisting of a backbone module (SRNs), a superpixel segmentation module, and a superpixel regularization module. In the proposed method, except for the novel view synthesis task, the task of representation-aware superpixel segmentation mask generation is realized by the proposed superpixel segmentation module. Then, the superpixel regularization module utilizes the superpixel segmentation mask to guide the backbone to be learned in a locally smooth way, and optimizes the scene representations of the local regions to indirectly alleviate the structure distortion of low-texture regions in a self-supervised manner. Extensive experimental results on both our constructed datasets and the public Synthetic-NeRF dataset demonstrated that the proposed SG-SRNs achieved a significantly better 3D structure representing performance.

We investigate the optimal control problem for networked control systems consisting of a linear plant controlled by a remote controller and a local controller. By virtue of the maximum principle, we establish a non-homogeneous relationship between the state and the costate of this class of systems. Based on this relationship, the optimal controllers are derived in terms of the two Riccati equations.

Fingerprinting based indoor positioning system is gaining more research interest under the umbrella of location-based services. However, existing works have certain limitations in addressing issues such as noisy measurements, high computational complexity, and poor generalization ability. In this work, a random vector functional link network based approach is introduced to address these issues. In the proposed system, a subset of informative features from many randomized noisy features is selected to both reduce the computational complexity and boost the generalization ability. Moreover, the feature selector and predictor are jointly learned iteratively in a single framework based on an augmented Lagrangian method. The proposed system is appealing as it can be naturally fit into parallel or distributed computing environment. Extensive real-world indoor localization experiments are conducted on users with smartphone devices and results demonstrate the superiority of the proposed method over the existing approaches.