We propose an online algorithm based on local sparse representation for robust object tracking. Local image patches of a target are represented by their codes with over-complete dictionary constructed online, and classifier is learned to discriminate the from background. To alleviate visual drift problem often encountered in tracking, two-stage proposed exploit both ground truth information first frame observations obtained online. Different recent discriminative tracking methods that use...

Despite the remarkable progress made by salient object detection of natural sensing images (NSI-SOD), complex background and scale diversity issues remote (RSIs) still pose a substantial obstacle. In this study, we build an end-to-end channel-enhanced remodeling-based network (CRNet) for optical RSIs (ORSIs) to highlight objects through feature augmentation. First, backbone convolutional block is used suggest fundamental characteristics. Then, use channel enhance module (CEM) shallow...

This paper reviews and evaluates several state-of-the-art online object tracking algorithms. Notwithstanding decades of efforts, remains a challenging problem due to factors such as illumination, pose, scale, deformation, motion blur, noise, occlusion. To account for appearance change, most recent algorithms focus on robust representations effective state prediction. In this paper, we analyze the components each method identify their key roles in dealing with specific challenges, thereby...

The past decade has witnessed the rapid development of feature representation learning and distance metric learning, whereas two steps are often discussed separately. To explore their interaction, this work proposes an end-to-end framework called DARI, i.e. Distance And Representation Integration, validates effectiveness DARI in challenging task person verification. Given training images annotated with labels, we first produce a large number triplet units, each one contains three images,...

Crowd counting, aiming at estimating the total number of people in unconstrained crowded scenes, has increasingly received attention. But it is greatly challenged by huge variation scale. In this paper, we propose a novel Multi-View Scale Aggregation Network (MVSAN), which handle scale from feature, input and criterion view comprehensively. Firstly, design simple but effective Multi-Scale Feature Encoder, exploits dilated convolution layers with various dilation rates to improve...

<div>To address the issues of unreasonable collision avoidance path planning algorithms and inadequate safety in high-speed scenarios, a trajectory prediction-based algorithm has been proposed. First, prediction model is constructed using long–short-term memory (LSTM) network, trained tested with HighD dataset. Second, future obstacle car predicted, information two cars combined to generate lane-changing decision, three-times B-spline curves are used clusters. The optimal paths...

The accurate characterization of wear debris is crucial for assessing the health rotating engine components and conducting simulation experiments in detection. This study proposed an intelligent recognition method ferrography images, leveraging several improved Mask Region-based Convolutional Neural Network (Mask R-CNN) algorithms to quantitatively calculate both number particles their coverage areas. improvement on R-CNN focuses two key aspects: enhancing feature extraction through pyramid...

How to avoid the invading of attack in biometric system, such as 2D printed photos, gradually becomes an important research hotspot. In this paper, we present a novel descriptor light field tackle issue. Based on angular and spatial information field, proposed histogram gradient (LFHoG) is derived from three directions, including vertical, horizontal depth. Different with traditional HoG image, depth direction distinctive field. To validate effectiveness LFHoG descriptor, experiments have...

This paper proposes a robust tracking algorithm by third-order tensor representation and adaptive appearance modeling. In this method, the target in each video frame is represented tensor. preserves spatial correlation inside region can integrate multiple cues for description. Based on representation, multilinear subspace learned online to model variations during tracking. Compared other methods, our approach detect local structure space fuse information from different feature spaces....

We present a novel algorithm that exploits joint optimization of representation and classification for robust tracking in which the goal is to minimize least-squares reconstruction errors discriminative penalties with regularized constraints. In this formulation, an object represented by sparse coefficients local patches based on overcomplete dictionary, classifier learned discriminate target from background. To locate each frame, we propose deterministic approach solve problem. show...

The counting task, which plays a fundamental role in numerous applications (e.g., crowd counting, traffic statistics), aims to predict the number of objects with various densities. Existing object tasks are designed for single class. However, it is inevitable encounter newly coming data new classes our real world. We name this scenario as evolving counting. In paper, we build first dataset and propose unified network attempt address task. proposed consists two key components: class-agnostic...

To aim at the problem of many researchers have only focused on recovering 3D human body information from color images, which is not accurate, causing great ambiguity and slow. we propose a new method for pose estimation. We get images depth through RGBD camera. use convolutional neural networks 2D estimation to joint points coordinates in image then map returned results corresponding obtain information. For estimation, improve accuracy stacked hourglass network using Faster-RCNN residual...

The extraction of human pose from the 3D point-cloud data is an important research challenge in field behavior recognition, which will play a vital role smart home, elderly-care, gaming, etc. Among variety technologies proposed, mmWave sensor based technology has been attracting more attention recognition due to its advantages privacy protection. However, previous studies have focused on using sophisticated radars (which are typically costly), whereas commodity sensors not received much...

This paper proposes a robust visual tracking approach based on saliency selection. In this method, salient patches and their spatial context inside the object region are exploited for representation appearance modeling. Tracking is then implemented by hybrid stochastic deterministic mechanism, which needs small number of samples particle filtering escapes local minimum in conventional tracking. As time progresses, selected updated online to adapt model both environmental changes. We carry...

In this paper, we propose a patch-based object tracking algorithm which provides both good enough robustness and computational efficiency. Our learns maintains Composite Patch-based Templates (CPT) of the target. Each composite template employs HOG, CS-LBP, color histogram to represent local statistics edges, texture flatness. The CPT model is initially established by maximizing discriminability templates given first frame, automatically updated on-line adding new effective patches deleting...

Appearance variation is a big challenge for object tracking. To deal with this problem, we propose robust tracking method by online appearance modeling and sparse representation. In method, use the intensity matrix of image to represent object, learn low dimensional subspace model variations during Then applying recent theory representation [1], construct likelihood function measure similarity between an candidate learned model. After that, led Bayesian inference framework, in which particle...

A new approach of tracking objects in image sequences is proposed, which the constant changes size and orientation target can be precisely described. For each incoming frame, a probability distribution created, where target's area turns into blob. The scale this blob determined based on local maxima differential scale-space filters. We employ QP_TR trust region algorithm to search orientational multi-scale normalized Laplacian filter locate as well determine its orientation. Based results...

The counting task, which plays a fundamental role in numerous applications (e.g., crowd counting, traffic statistics), aims to predict the number of objects with various densities. Existing object tasks are designed for single class. However, it is inevitable encounter newly coming data new classes our real world. We name this scenario as \textit{evolving counting}. In paper, we build first evolving dataset and propose unified network attempt address task. proposed model consists two key...