A5101882875- Video Surveillance and Tracking Methods
- Advanced Vision and Imaging
- Advanced Image Processing Techniques
- Human Pose and Action Recognition
- Anomaly Detection Techniques and Applications
- Advanced Image and Video Retrieval Techniques
- Image Enhancement Techniques
- Image Processing Techniques and Applications
- Image and Video Quality Assessment
- Image and Signal Denoising Methods
- Visual Attention and Saliency Detection
- Video Analysis and Summarization
- Topic Modeling
- Data Management and Algorithms
- Advanced Neural Network Applications
- Gait Recognition and Analysis
- Video Coding and Compression Technologies
- Expert finding and Q&A systems
- Face and Expression Recognition
- Recommender Systems and Techniques
- Machine Learning and Data Classification
- Advanced Measurement and Detection Methods
- Caching and Content Delivery
- Mobile Crowdsensing and Crowdsourcing
- Domain Adaptation and Few-Shot Learning
Guilin University of Electronic Technology
2024
Beijing Institute of Technology
2014-2024
Shihezi University
2024
Beijing University of Posts and Telecommunications
2011-2024
Shenzhen MSU-BIT University
2024
Shenzhen University
2024
Microsoft Research (United Kingdom)
2020-2022
Zhejiang University of Technology
2022
Australian National University
2021
Beijing Haidian Hospital
2021
Occlusions pose a significant challenge to optical flow algorithms that rely on local evidences. We consider an occluded point be one is imaged in the reference frame but not next, slight overloading of standard definition since it also includes points move out-of-frame. Estimating motion these extremely difficult, particularly two-frame setting. Previous work relies CNNs learn occlusions, without much success, or requires multiple frames reason about occlusions using temporal smoothness. In...
We perform fast vehicle detection from traffic surveillance cameras. A novel deep learning framework, namely Evolving Boxes, is developed that proposes and refines the object boxes under different feature representations. Specifically, our framework embedded with a light-weight proposal network to generate initial anchor as well early discard unlikely regions; fine-turning produces detailed features for these candidate boxes. show intriguingly by applying fusion techniques, can be refined...
Classic query optimization techniques, including predicate pushdown, are of limited use for machine learning inference queries, because the user-defined functions (UDFs) which extract relational columns from unstructured inputs often very expensive; predicates will remain stuck behind these UDFs if they happen to require that generated by UDFs. In this work, we demonstrate constructing and applying probabilistic filter data blobs do not satisfy predicate; such filtering is parametrized...
The rapid advances of transportation infrastructure have led to a dramatic increase in the demand for smart systems capable monitoring traffic and street safety. Fundamental these applications are community-based evaluation platform benchmark object detection multi-object tracking. To this end, we organize AVSS2017 Challenge on Advanced Traffic Monitoring, conjunction with International Workshop Street Surveillance Safety Security (IWT4S), evaluate state-of-the-art tracking algorithms...
Triggered by several head-mounted display (HMD) devices that have come to the market recently, such as Oculus Rift, HTC Vive, and Samsung Gear VR, significant interest has developed in virtual reality (VR) systems, experiences applications. However, current HMD are still very heavy large, negatively affecting user experience. Moreover, VR approaches perform rendering locally either on a mobile device tethered an HMD, or computer/console HMD. In this paper, we discuss how enable truly...
Unbalanced interaction relationships at personal and group levels play a pivotal role in collective activity recognition, which has not been adaptively jointly explored by previous approaches. In this paper, we propose graph attention model (GAIM) embedded with the block (GAB) to explicitly infer unbalanced relations unified architecture, further learn spatial temporal evolutions of from these interactions predict labels. We first design spatiotemporal graphs tailored where concurrent person...
The robust tracking of abrupt motion is a challenging task in computer vision due to its large uncertainty. While various particle filters and conventional Markov-chain Monte Carlo (MCMC) methods have been proposed for visual tracking, these often suffer from the well-known local-trap problem or poor convergence rate. In this paper, we propose novel sampling-based scheme Bayesian filtering framework. To effectively handle problem, first introduce stochastic approximation (SAMC) sampling...
Vehicle tracking is an important part in intelligent transportation surveillance. But now vehicle faces with the problems such as scale change, interference of similar color, low resolution video data and so on. In this paper improved Markov chain Monte Carlo(MCMC) named optical flow MCMC(OF-MCMC) sampling algorithm proposed for tracking. First, we use method to get moving direction initial frames, which can solve problem what's more speed replaces second-order autoregressive motion model...
State-of-the-art neural network models estimate large displacement optical flow in multi-resolution and use warping to propagate the estimation between two resolutions. Despite their impressive results, it is known that there are problems with approach. First, of fails situations where small objects move fast. Second, creates artifacts when occlusion or dis-occlusion happens. In this paper, we propose a new module, Deformable Cost Volume, which alleviates problems. Based on designed Volume...
Locating actions in long untrimmed videos has been a challenging problem video content analysis. The performances of existing action localization approaches remain unsatisfactory precisely determining the beginning and end an action. Imitating human perception procedure with observations refinements, we propose novel three-phase framework. Our framework is embedded Actionness Network to generate initial proposals through frame-wise similarity grouping, then Refinement conduct boundary...
Road crack detection is of paramount importance for ensuring vehicular traffic safety, and implementing traditional methods cracks inevitably impedes the optimal functioning traffic. In light above, we propose a USSC-YOLO-based target algorithm unmanned aerial vehicle (UAV) road based on machine vision. The aims to achieve high-precision at all scale levels. Compared with original YOLOv5s, main improvements USSC-YOLO are ShuffleNet V2 block, coordinate attention (CA) mechanism, Swin...
Robust tracking of abrupt motion is a challenging task in computer vision due to the large uncertainty. In this paper, we propose stochastic approximation Monte Carlo (SAMC) based scheme for problem Bayesian filtering framework. our scheme, particle weight dynamically estimated by learning density states simulations, and thus local-trap suffered conventional MCMC sampling-based methods could be essentially avoided. addition, design an adaptive SAMC sampling method further speed up process...
Cloud gaming allows games to be rendered on the cloud server and videos encoded streamed in real time player's devices. Compared with other video streaming applications, offers a unique opportunity enhance encoding process by exploiting rendering information. In this paper, we propose two techniques improve encoding, aiming at enhancing perceived quality reducing computational complexity, respectively. First, develop rendering-based prioritized technique game according network bandwidth...
Spammer detection on social network is a challenging problem. The rigid anti-spam rules have resulted in emergence of "smart" spammers. They resemble legitimate users who are difficult to identify. In this paper, we present novel spammer classification approach based Latent Dirichlet Allocation(LDA), topic model. Our extracts both the local and global information distribution patterns, which capture essence spamming. Tested one benchmark dataset self-collected dataset, our proposed method...
In this letter, we present a new algorithm for single image super-resolution using the analysis sparse prior in <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">l</i> αβ color space. Experimental results show that our outperforms other existing state-of-the-art methods. addition, due to high scalability of algorithm, key modules proposed can be integrated with super resolution algorithms.
Interactive video segmentation systems aim at producing sub-pixel-level object boundaries for visual effect applications. Recent approaches mainly focus on using sparse user input (i.e. scribbles) efficient segmentation, however, the quality of final is not satisfactory following reasons: (1) boundary each frame often accurate, (2) across adjacent frames wiggle around inconsistently, causing temporal flickering, and (3) there a lack direct control fine tuning. We propose Coherent Parametric...
This paper presents a novel discriminative, generative, and collaborative appearance model for robust object tracking. In contrast to existing methods, we use different manifolds represent the target in discriminative generative models propose scheme combine these two components. particular: 1) component, develop graph regularized discriminant analysis (GRDA) algorithm that can find projection more effectively distinguish from background; 2) introduce simple yet effective coding method...
We consider the problem of pre-training models which convert structured datasets into succinct summaries that can be used to answer cardinality estimation queries. Doing so avoids per-dataset training and, in our experiments, reduces time construct by up 100×. When change, are incrementally updateable. Our key insights use multiple per dataset, learned for columnsets other simpler techniques do not achieve high accuracy, and analogous similar pre-trained images text, have some common...
We consider accelerating machine learning (ML) inference queries on unstructured datasets. Expensive operators such as feature extractors and classifiers are deployed user-defined functions (UDFs), which not penetrable with classic query optimization techniques predicate push-down. Recent schemes (e.g., Probabilistic Predicates or PP) assume independence among the predicates, build a proxy model for each offline, rewrite new by injecting these cheap models in front of expensive ML UDFs. In...