Various types of video can be captured with fisheye lenses; their wide field view is particularly suited to surveillance video. However, lenses introduce distortion, and this changes as objects in the scene move, making difficult interpret. Current still image correction methods are either limited small angles view, or strongly content dependent, therefore unsuitable for processing streams. We present an efficient robust scheme correction, which minimizes time-varying distortion preserves...

A multi-view object detection approach based on deep learning is proposed in this paper. Classical methods regression models are introduced, and the reasons for their weak ability to detect small objects analyzed. To improve performance of these methods, a proposed, model structure working principles explained. Additionally, retrieval accuracy both corresponding classical evaluated compared test dataset. The experimental results show that terms capability, Multi-view YOLO (You Only Look...

To address the poor image feature extraction ability, excessive tracking time, and low accuracy of traditional real-time algorithms for aerial vehicle traffic, a algorithm using improved convolutional neural network (CNN) transfer learning was proposed herein. In this algorithm, first, images are matched with sample offset calibration. Second, CNN parameters initialized by constructing filter set, is employed to construct pre-training model. Third, deep convolution structure map designed...

Existing scene text spotters are designed to locate and transcribe texts from images. However, it is challenging for a spotter achieve precise detection recognition of simultaneously. Inspired by the glimpse-focus spotting pipeline human beings impressive performances Pre-trained Language Models (PLMs) on visual tasks, we ask: 1) "Can machines spot without just like beings?", if yes, 2) "Is block another alternative other than word or character?" To this end, our proposed leverages advanced...

This essay designed the image acquisition and processing system of color sorter based on FPGA. It completed construction hardware platform design algorithms. In workflow, firstly collects information through FPGA-controlled CCD sensor, then processes data After processing, FPGA outputs recognition results to blowing control module complete screening function fast moving objects. To test design, simulation were carried out at end essay. The showed that had obviously improved purity products.

The routing algorithm of a Network-on-Chip (NoC) is essential to its performance and power consumption.This paper presents novel shortest path for TriBA-Net.First, the designs coding scheme based on topological features TriBA-Net.The set words 1, 3, 2, used in scheme, has same meaning as well-known group S3 3-letters.Second, communication model, which contains 6 types flow modes, been proposed reflecting status within two hops.Finally, simplified by cyclic permutation characteristic...

Data valuation is essential for quantifying data's worth, aiding in assessing data quality and determining fair compensation. While existing methods have proven effective evaluating the value of Euclidean data, they face limitations when applied to increasingly popular graph-structured data. Particularly, graph introduces unique challenges, primarily stemming from intricate dependencies among nodes exponential growth estimation costs. To address challenging problem valuation, we put forth an...

Innovative processor architecture designs are shifting towards Many-Core Architectures (MCAs) to meet the future demands of high-performance computing as limits Moore’s Law have almost been reached. Many-core processors utilize shared memory hierarchies achieve high-speed systems, improving access efficiency. However, number cores multiplies, scalability this system is significantly constrained by increased proportion long-distance and Non-Uniform Memory Access (NUMA). Improving MCAs crucial...

Many multimedia analysis algorithms rely on probability distributions that characterize audio or image features as generally high dimensions.For example, music methods, such automatic transcription (AMT) [1] and classification [2], in these applications, having sufficient similarity (or equivalent difference) between becomes crucial.The classical distance difference of density includes Kullback Leibler divergence, Kolmogorov distance, Bhattacharyya (also known Hellinger distance),...

In the field of substation operation and maintenance, recognizing behaviors maintenance personnel is crucial for maintaining safe power grid ensuring safety workers. To enhance accuracy human behavior recognition in this paper proposes an ECA-MSG3D algorithm model work apparatus based on Multi-Scale Pointwise Convolutional Neural Network 3D Object Detection (MSG3D) model. The MSG3D module, which utilizes a multi-scale aggregation scheme unified spatiotemporal graph convolution, used feature...

This paper focuses on the critical address aligning issues in real time DSP image processing system. SAD and byte array moving operation which are quite sensitive to discussed thoroughly, get well solved. At end of paper, we present an optimized design implementation template matching system based high performance TigerSHARC COTS module solution severe issues.

The paper proposes a visual tracking method based on deep learning without online model updating. In consideration of the advantages in feature representation, SSD (Single Shot Multibox Detector) is used as object extractor model. Simultaneously, color histogram and HOG (Histogram Oriented Gradient) are combined to select object. process tracking, multi-scale searching map built improve detection performance efficiency. experiment eight respective video sequences baseline dataset, compared...

This paper focuses on resource access issue for the large scale COTS based radar processing systems. Since designing and deployment of an agent in system is commonly accepted approach, several major schemes are inspected discussed detail. Then we reach point that a hybrid architecture more feasible At end paper, practical solution provided proved to be scalable effective. (4 pages)

This paper focuses on the monitoring and debugging issues of concurrent DSP processing system in real time large scale distributed parallel scenario. It strives for putting record replay technique which stems from computing realm, into practice modern high performance embedded (HPEC) application enhancing its feasibility scalability. Architecture event recording based COTS modules is established, core components are designed implemented, including global clocking domain, generator recorder.

In the image super-resolution reconstruction, many methods based on deep learning mostly adopt traditional mean squared error (MSE) as loss function, and reconstructed is prone to problem of fuzzy details too smooth. order solve this problem, paper improves square function proposes an reconstruction method multi-scale feature function. The whole network model consists a DenseNet-based convolutional neural which used optimize Taking corresponding original HD input convolved in series,...

Abstract In order to improve the performance of automatic classifier for ground targets under complex background, a new which was based on bag words model and support vector machine (SVM) build. Firstly, keypoints target image were extracted into vectors by using Scale-Invariant Feature Transform algorithm, from all trained images merged large vectors. Secondly, clustered small included K -means clustering algorithm. To analyze how different values affected classifying results, set several...