A5100386594- Human Pose and Action Recognition
- Video Surveillance and Tracking Methods
- Gait Recognition and Analysis
- Anomaly Detection Techniques and Applications
- Human Motion and Animation
- Advanced Neural Network Applications
- Video Analysis and Summarization
- Hand Gesture Recognition Systems
- Domain Adaptation and Few-Shot Learning
- Advanced Image and Video Retrieval Techniques
- Medical Image Segmentation Techniques
- Generative Adversarial Networks and Image Synthesis
- Adversarial Robustness in Machine Learning
- Brain Tumor Detection and Classification
- Digital Media and Visual Art
- Supply Chain and Inventory Management
- Data Management and Algorithms
- Geographic Information Systems Studies
- Advanced Computing and Algorithms
- Computational and Text Analysis Methods
- Gaze Tracking and Assistive Technology
- Influenza Virus Research Studies
- Advanced Glycation End Products research
- Sleep and Work-Related Fatigue
- Organizational Management and Leadership
Beijing University of Posts and Telecommunications
2018-2024
Northeast Agricultural University
2023-2024
Hebei University of Engineering
2022-2023
Alibaba Group (China)
2023
Wuxi Institute of Arts & Technology
2023
China Earthquake Administration
2021-2022
Qiongtai Teachers College
2022
Institute of Disaster Prevention
2021
National University of Singapore
2020
Chinese Academy of Surveying and Mapping
2020
Human motion prediction is an increasingly interesting topic in computer vision and robotics. In this paper, we propose a new end-to-end feedforward network, TrajectoryCNN, to predict future poses. Compared with the most existing methods, introduce trajectory space focus on modeling dynamics of input sequence coupled spatio-temporal features, dynamic local-global global temporal co-occurrence features space. Specifically, describe spatial structural information hidden natural human sequence,...
Consistency regularization and pseudo labeling-based semi-supervised methods perform co-training using the labels from multi-view inputs. However, such models tend to converge early a consensus, degenerating self-training ones, produce low-confidence perturbed inputs during training. To address these issues, we propose an Uncertainty-guided Collaborative Mean-Teacher (UCMT) for semantic segmentation with high-confidence labels. Concretely, UCMT consists of two main components: 1)...
In order to address the shortcomings of traditional bidirectional RRT* algorithm, such as its high degree randomness, low search efficiency, and many inflection points in planned path, we institute improvements following directions. Firstly, problem randomness process random tree expansion, expansion direction growing at starting point is constrained by improved artificial potential field method; thus, grows towards target point. Secondly, sampling grown biased number Finally, path algorithm...
Human motion prediction is challenging due to the complex spatiotemporal feature modeling. Among all methods, graph convolution networks (GCNs) are extensively utilized because of their superiority in explicit connection Within a GCN, correlation adjacency matrix drives aggregation and key extracting predictive features. State-of-the-art methods decompose into spatial correlations for each frame temporal joint. Directly parameterizing these introduces redundant parameters represent common...
High throughput and low latency inference of deep neural networks are critical for the deployment learning applications. This paper presents efficient techniques IntelCaffe, first Intel(R) optimized framework that supports 8-bit precision model optimization convolutional on Xeon(R) Scalable Processors. The is automatically generated with a calibration process from FP32 without need fine-tuning or retraining. We show ResNet-50, Inception-v3 SSD improved by 1.38X-2.9X 1.35X-3X respectively...
Paleoseismic offsets are important parameters for evaluating fault activity. With the development and popularization of unmanned aerial vehicles (UAVs) structure-from-motion (SfM) photogrammetry, more low-cost mini-UAVs have been used geoscience studies like active faults paleoearthquakes. In this study, we take Gebi ridge in middle Altyn Tagh (ATF) western China as an example using control-free images acquired by a mini-UAV SfM to measure paleoseismic offsets. The measurement accuracies...
Point of interest (POI) matching is critical but the most technically difficult part multi-source POI fusion. The accurate POIs from different sources important for effective reuse data. However, existing research on usually adopts weak constraints, which leads to a low accuracy. To address shortcomings previous studies, this paper proposes method with multiple determination constraints. First, according various attributes (name, class, and spatial location), new calculation model...
Topic extraction and evolution analysis became a research hotspot in the academic community due to its ability reveal development trend of certain field discover law topic content different stages field. However, current methods still face challenges, such as inaccurate recognition unclear paths, which can seriously compromise comprehensiveness accuracy analysis. To address problem, paper proposes path method based on LDA2vec symmetry model. Under given conditions, both LDA Word2vec used...
Action recognition is a significant and challenging topic in the field of sensor computer vision. Two-stream convolutional neural networks (CNNs) 3D CNNs are two mainstream deep learning architectures for video action recognition. To combine them into one framework to further improve performance, we proposed novel network, named spatiotemporal interaction residual network with pseudo3D (STINP). The STINP possesses three advantages. First, consists branches constructed based on (ResNets)...
Investigating post-earthquake surface ruptures is important for understanding the tectonics of seismogenic faults. The use unmanned aerial vehicle (UAV) images to identify has advantages low cost, fast data acquisition, and high processing efficiency. With rapid development deep learning in recent years, researchers have begun using it image crack detection. However, due complex background diverse characteristics ruptures, remains challenging quickly train an effective automatic earthquake...
Amodal segmentation is a new direction of instance while considering the visible and occluded parts instance. The existing state-of-the-art method uses multi-task branches to predict amodal part separately subtract from obtain part. However, contains information. Therefore, separated prediction will generate duplicate Different this method, we propose based on idea jigsaw. two naturally decoupled occluded, which like getting matching jigsaw pieces. Then put pieces together get This makes...
Human motion prediction is an important and challenging task in computer vision with various applications. Recurrent neural networks (RNNs) convolutional (CNNs) have been proposed to address this task. However, RNNs exhibit their limitations on long-term temporal modelling spatial of signals. CNNs show inflexible capability that mainly depends a large kernel the stride operation. Moreover, those methods predict multiple future poses recursively, which easily suffer from noise accumulation....
In this paper a technique for detecting the daily fatigue using characteristics of local facial image is discussed. To reduce influence environment illumination and retain more details features, images are preprocessed by method color consistent area correction. The texture features derived from gray level co-occurrence matrix two kinds space, reliability universality system improved difference algorithm. Then error backward propagation algorithm employed to form complete detection Finally,...
In many practical engineering applications, the number of actions that have been finished should be known, particularly for an untrimmed video sequence includes event with a series actions, it is important to know finished. this paper, we termed process as visual progress estimation (EPE). However, research related problem few in community. To solve problem, human action analysis-based framework, namely one-shot simultaneously detection and identification (SADI)-EPE, presented paper. The EPE...
Pose prediction is to predict future poses given a window of previous poses. In this paper, we propose new problem that predicts using 3D joint coordinate sequences. Different from the traditional pose based on Mocap frames, convenient use in real applications due its simple sensors capture data. We also present framework, PISEP^2 (Pseudo Image Sequence Evolution Prediction), address problem. Specifically, skeletal representation proposed by transforming sequence into an image sequence,...
Predicting future human motion is critical for intelligent robots to interact with humans in the real world, and has nature of multi-granularity. However, most existing work either implicitly modeled multi-granularity information via fixed modes or focused on modeling a single granularity, making it hard well capture this accurate predictions. In contrast, we propose novel end-to-end network, Semi-Decoupled Multi-grained Trajectory Learning network (SDMTL), predict poses, which not only...
Action prediction is an important task in human activity analysis, which has many practical applications, such as human–robot interactions and autonomous driving. often comprises two subtasks: action semantic future motion prediction. Most of the existing works treat these subtasks separately, ignoring correlations, leading to unsatisfying performance. By contrast, we jointly model tasks improve predictions utilizing their semantics. In terms methodology, propose a novel multi-task framework...
The recent advancements in 2D generation technology have sparked a widespread discussion on using priors for 3D shape and texture content generation. However, these methods often overlook the subsequent user operations, such as aliasing blurring that occur when acquires model simplifies its structure. Traditional graphics partially alleviate this issue, but synthesis technologies fail to ensure consistency with original model's appearance cannot achieve high-fidelity restoration. Moreover,...