A5101633987- Human Pose and Action Recognition
- Hand Gesture Recognition Systems
- Video Surveillance and Tracking Methods
- Metaheuristic Optimization Algorithms Research
- Neural Networks and Applications
- Tactile and Sensory Interactions
- Multimodal Machine Learning Applications
- Traffic Prediction and Management Techniques
- Anomaly Detection Techniques and Applications
- Video Analysis and Summarization
- Gaze Tracking and Assistive Technology
- Face and Expression Recognition
- Image Enhancement Techniques
- Advanced Image Fusion Techniques
- Advanced Neural Network Applications
- Advanced Multi-Objective Optimization Algorithms
- Teleoperation and Haptic Systems
- Advanced Data Compression Techniques
- Mobile Crowdsensing and Crowdsourcing
- Image Processing and 3D Reconstruction
- Neural dynamics and brain function
- Data Management and Algorithms
- Context-Aware Activity Recognition Systems
- Expert finding and Q&A systems
- Modular Robots and Swarm Intelligence
China Mobile (China)
2023
Tianshui Normal University
2022
Baidu (China)
2021
Shenzhen Institutes of Advanced Technology
2021
Chinese Academy of Sciences
2021
Institute of Information Science, Academia Sinica
2011
Ta Hwa University of Science and Technology
2009
National Central University
2004-2007
Tamkang University
2001-2003
Long-range and short-range temporal modeling are two complementary crucial aspects of video recognition. Most the state-of-the-arts focus on spatio-temporal then average multiple snippet-level predictions to yield final video-level prediction. Thus, their prediction does not consider features how evolves along dimension. In this paper, we introduce a novel Dynamic Segment Aggregation (DSA) module capture relationship among snippets. To be more specific, attempt generate dynamic kernel for...
The detection of traffic anomalies is a critical component the intelligent city transportation management sys-tem. Previous works have proposed variety notable in-sights and taken step forward in this field, however, dealing with complex environment remains challenge. Moreover, lack high-quality data complexity scene, motivate us to study problem from hand-crafted perspective. In paper, we propose straightforward efficient framework that includes pre-processing, dynamic track module,...
A new approach to optimization problems based on the self-organizing feature maps is proposed. We name algorithm SOM-based (SOMO) algorithm. Through process, good solutions an problem can be simultaneously explored and exploited. An additional advantage of that outputs neural network allow us transform a multi-dimensional fitness landscape into three-dimensional projected landscape. Several simulations are used illustrate effectiveness proposed
One particular application of gesture-based systems is to implement a speaking aid for the deaf. For this happen, it requires module that can recognize three-dimensional (3-D) arm movements since movement one four main attributes characterizing sign word. In paper authors propose fuzzy rule-based recognition method. The effectiveness system evaluated by recognizing 3-D involved in Taiwanese Sign Language (TSL).
The detection of traffic anomalies is a critical component the intelligent city transportation management system. Previous works have proposed variety notable insights and taken step forward in this field, however, dealing with complex environment remains challenge. Moreover, lack high-quality data complexity scene, motivate us to study problem from hand-crafted perspective. In paper, we propose straightforward efficient framework that includes pre-processing, dynamic track module,...
In recent years, deep learning, powered by neural networks, has achieved widespread success in solving high-dimensional problems, particularly those with low-dimensional feature structures. This stems from their ability to identify and learn low dimensional features tailored the problems. Understanding how networks extract such during training dynamics remains a fundamental question learning theory. this work, we propose novel perspective interpreting neurons last hidden layer of network as...
In this paper, we present JanusNet, an efficient CNN model that can perform online background subtraction and robustly detect moving targets using resource-constrained computational hardware on-board unmanned aerial vehicles (UAVs). Most of the existing work on sub-traction either assume camera is stationary or make limiting assumptions about motion camera, structure scene under observation, apparent in video. JanusNet does not have these limitations therefore, applicable to a variety UAV...
The object of this paper is to present a set techniques integrated into two low-cost human computer interfaces. Although the interfaces have many potential applications, one main application help disabled persons attain or regain some degree independent communications and control. first interface voice-controlled mouse second an accelerometer-based mouse.
Support vector machine(SVM) can effectively solve the classification problem with small samples, nonlinear and high dimensions, but it exits weak generalization ability low accuracy. So an improved genetic algorithm(IGA) is introduced in order to propose a new classification(IGASVM) method based on combining GA SVM model. In proposed IGASVM method, self-adaptive control parameter strategy improving convergence speed are into keep diversity of population, promptly reflect premature individual...
Color quantization is a process of sampling three-dimensional color space (e.g. RGB) to reduce the number colors in image. By reducing discrete subset known as codebook or palette, each pixel original image mapped an entry according these palette colors. In this paper, reinforcement-learning approach proposed. Fuzzy rules, which are used select appropriate parameters for adaptive clustering algorithm applied quantization, built through reinforcement learning. comparing new method with on 30...
There are many different approaches to recognition of spatio-temporal patterns. Each has its own merits and disadvantages. In this paper we present a neural-network-based approach pattern recognition. The effectiveness method is evaluated by recognizing 3D arm movements involved in Taiwanese sign language.
Data-driven black-box model-based optimization (MBO) problems arise in a great number of practical application scenarios, where the goal is to find design over whole space maximizing target function based on static offline dataset. In this work, we consider more general but challenging MBO setting, named constrained (CoMBO), only part can be optimized while rest by environment. A new challenge arising from CoMBO that most observed designs satisfy constraints are mediocre evaluation....
The purpose of this study is to design a portable virtual piano. By utilizing optical fiber gloves and the piano software designed by study, user can play anywhere at any time. This consists three major parts: finger tapping identification, hand movement positioning MIDI sound effect simulation. To piano, wears simulates key motions. bending information detected tell when motions are made. Images captured video camera analyzed, locations moving directions positioned, corresponding scales...
Abstract In order to effectively remove the mixed noise, a novel algorithm for image reducing noise combined Pulse Coupled Neural Networks (PCNN) and regularization of Perona-Malik equation (P-M equation) is put forward. Firstly, positions impulse pollution are located by using PCNN, first stage adaptive filtering applied according result detection. Secondly, multi-direction information median local neighborhood noisy information. And then Gaussian in denoised utilizing P-M diffusion...
Long-range and short-range temporal modeling are two complementary crucial aspects of video recognition. Most the state-of-the-arts focus on spatio-temporal then average multiple snippet-level predictions to yield final video-level prediction. Thus, their prediction does not consider features how evolves along dimension. In this paper, we introduce a novel Dynamic Segment Aggregation (DSA) module capture relationship among snippets. To be more specific, attempt generate dynamic kernel for...