嵌入式影像處理系統及 SURF 特徵擷取

Due to the advancements in digital technologies and social networking, image collections are growing exponentially. The important aim in content-based image retrieval (CBIR) is to reduce the semantic gap issue that improves the performance of image retrieval. In this paper, the objective is achieved by introducing effective visual words fusion technique based on speeded-up robust features (SURF) and histograms of oriented gradients (HOG) feature descriptors. HOG is used to extract global features, whereas SURF is used for the extraction of local features. Global features are preferred for large-scale image retrieval, whereas local features perform better on those systems that support semantic queries with close visual appearance. Moreover, SURF is scale and rotation-invariant as compared to HOG descriptor and it works better for low illumination. On the contrary, HOG performs better for scene-recognition- or activity-recognition-based applications. In the proposed technique, visual words fusion of SURF and HOG feature descriptors is carried which performed better than features fusion of SURF and HOG feature descriptors as well as state-of-the-art CBIR techniques. The proposed technique based on visual words fusion gives classification accuracy of 98.40% using support vector machine while image retrieval accuracy of 80.61%. Qualitative and quantitative analyses performed on four standard image collections namely, Corel-1000, Corel-1500, Corel-5000, and Caltech-256 demonstrate the effectiveness of the proposed technique based on visual words fusion of SURF and HOG feature descriptors.

Motion estimation in videos is a computationally intensive process. A popular strategy for dealing with such a high processing load is to accelerate algorithms with dedicated hardware such as graphic processor units (GPU), field programmable gate arrays (FPGA), and digital signal processors (DSP). Previous approaches addressed the problem using accelerators together with a general purpose processor, such as acorn RISC machines (ARM). In this work, we present a co-processing architecture using FPGA and DSP. A portable platform for motion estimation based on sparse feature point detection and tracking is developed for real-time embedded systems and smart video sensors applications. A Harris corner detection IP core is designed with a customized fine grain pipeline on a Virtex-4 FPGA. The detected feature points are then tracked using the Lucas---Kanade algorithm in a DSP that acts as a co-processor for the FPGA. The hybrid system offers a throughput of 160 frames per second (fps) for VGA image resolution. We have also tested the benefits of our proposed solution (FPGA + DSP) in comparison with two other traditional architectures and co-processing strategies: hybrid ARM + DSP and DSP only. The proposed FPGA + DSP system offers a speedup of about 20 times and 3 times over ARM + DSP and DSP only configurations, respectively. A comparison of the Harris feature detection algorithm performance between different embedded processors (DSP, ARM, and FPGA) reveals that the DSP offers the best performance when scaling up from QVGA to VGA resolutions.

Object recognition is a key research area in the field of image processing and computer vision, which recognizes the object in an image and provides a proper label. In the paper, three popular feature descriptor algorithms that are Scale Invariant Feature Transform (SIFT), Speeded Up Robust Feature (SURF) and Oriented Fast and Rotated BRIEF (ORB) are used for experimental work of an object recognition system. A comparison among these three descriptors is exhibited in the paper by determining them individually and with different combinations of these three methodologies. The amount of the features extracted using these feature extraction methods are further reduced using a feature selection (k-means clustering) and a dimensionality reduction method (Locality Preserving Projection). Various classifiers i.e. K-Nearest Neighbor, Naive Bayes, Decision Tree, and Random Forest are used to classify objects based on their similarity. The focus of this article is to present a study of the performance comparison among these three feature extraction methods, particularly when their combination derives in recognizing the object more efficiently. In this paper, the authors have presented a comparative analysis view among various feature descriptors algorithms and classification models for 2D object recognition. The Caltech-101 public dataset is considered in this article for experimental work. The experiment reveals that a hybridization of SIFT, SURF and ORB method with Random Forest classification model accomplishes the best results as compared to other state-of-the-art work. The comparative analysis has been presented in terms of recognition accuracy, True Positive Rate (TPR), False Positive Rate (FPR), and Area Under Curve (AUC) parameters.

The stereo matching algorithm of binocular underwater images based on SURF (Speeded Up Robust Features) operator and epipolar restriction is presented. With complex underwater environment, epipolar restriction is introduced to solve the impact of refraction. Fast hessian-matrix detector is used to detect the interest points and the SURF descriptor is used to describe the points’ features. Euclidean distance and epipolar restriction are used for features matching to simplify the calculation. The method has better performance than traditional algorithm on computing efficiency and matching accuracy.

Indexing images represents a necessary tool for various domains related to computer vision such as video surveillance and movements analysis. Visual characteristics of image such as color, texture, shape are used to identify the content of these frames. The high increase of data (Big Data) makes us face a problem of curse of dimensionality. In this paper, we propose a cloud-based technique for images indexing images using PCA (Principal component analysis) and a binary tree. Our work consists of exploiting SIFT (Scale Invariant Feature Transform) and SURF (Speeded up Robust Features) descriptors as frames features, PCA as dimensionality reduction method of SIFT and SURF features and the binary tree representation of these images in order to accelerate the research phase and reduce the storage space. Our approach consists of developing a system based on three phases: indexing images with SIFT and SURF features, compression within the application of PCA and finally the image retrieval in the cloud.

This paper presents the image processing algorithm for wireless charging between each mobile robot. The image processing algorithm converts Red Green Blue (RGB) format of inputted image to detect edge. It calculates a specific area using Hough Transformation (HT) in detected edge and judges correct charging antenna using Speeded-Up Robust Features (SURF). Accordingly, the image processing algorithm can control position and direction of mobile robot and antenna for wireless charging. The image processing algorithm is implemented wireless charging systems, which are set up on each two mobile robot and it is verified with experiment.

Speeded Up Robust Features (SURF) algorithm is a fast robust local feature matching method. Because of the advantage of invariance of scale and rotation, the SURF algorithm has been widely used in the image registration. However, due to the acquired angle, imaging mode, different resolution, SURF algorithm becomes quite difficult in the image registration. This paper presents a normalized SURF algorithm which can reduce the impact of hue difference between remote sensing images. Combined with the idea of two-way matching, the normalized SURF is able to improve the accuracy of the image registration. The experimental results show the normalized SURF algorithm can keep greater accuracy and relatively higher speed of matching than the original SURF algorithm.

The combination of hardware acceleration and flexibility make FPGAs (field programmable gate arrays) important to image processing applications. There is also a need for efficient, flexible hardware/software codesign environments that can balance the benefits and costs of using FPGAs. Image processing applications often consist of pipeline of components where each component applies a different processing algorithm. Components can be implemented for FPGAs or software. Such systems enable an image analyst to work with either FPGA or software implementations of image processing algorithms for a given problem. The pipeline assignment problem chooses from alternative implementations of pipeline components to yield the fastest pipeline. Our codesign system solves the pipeline assignment problem to provide the most effective implementation automatically, so the image analyst can focus solely on choosing components, which make up the pipeline. However, the pipeline assignment problem is NP complete. An efficient, dynamic solution to the pipeline assignment problem is a desirable enabler of codesign systems which use both FPGA and software implementations. This paper is concerned with solving pipeline assignment in this context. Consequently, we focus on optimal and heuristic methods for fast (fixed time limit) runtime pipeline assignment are investigated. We present experimental finding for pipelines of twenty or fewer components, which show that in our environment, optimal runtime solutions are possible for smaller pipelines and nearly optimal heuristic solutions are possible for larger pipelines.

The recently proposed, KAZE image feature detection and description algorithm (Alcantarilla et al. in Proceedings of the British machine vision conference. LNCS, vol 7577, no 6, pp 13.1–13.11, 2013) offers significantly improved robustness in comparison to conventional algorithms like SIFT (scale-invariant feature transform) and SURF (speeded-up robust features). The improved robustness comes at a significant computational cost, however, limiting its use for many applications. We report a GPU acceleration of the KAZE algorithm that is significantly faster than its CPU counterpart. Unlike previous reports, our acceleration does not resort to binary descriptors and can serve as a drop-in replacement for CPU-KAZE, SIFT, SURF etc. By achieving nearly tenfold speedup (for a 1920 by 1200 sized image, our Compute Unified Device Architecture (CUDA)-C implementation took around 245 ms on a single GPU in comparison to nearly 2400 ms for a 16-threaded CPU version) without degradation in feature extraction performance, our work expands the applicability of the KAZE algorithm. Additionally, the strategies described here could also prove useful for the GPU implementation of other nonlinear scale-space-based image processing algorithms.

Machine vision can prevent additional stress on yarn caused by contact measurement, as well as the risk of hairiness and breakage. However, the speed of the machine vision system is limited by image processing, and the tension detection method based on the axially moving model does not take into account the disturbance on yarn caused by motor vibrations. Thus, an embedded system combining machine vision with a tension observer is proposed. The differential equation for the transverse dynamics of the string is established using Hamilton's principle and then solved. A field-programmable gate array (FPGA) is used for image data acquisition, and the image processing algorithm is implemented using a multi-core digital signal processor (DSP). To obtain the yarn vibration frequency in the axially moving model, the brightest centreline grey value of the yarn image is put forward as a reference to determine the feature line. The calculated yarn tension value is then combined with the value obtained using the tension observer based on an adaptive weighted data fusion method in a programmable logic controller (PLC). The results show that the accuracy of the combined tension is improved compared with the original two non-contact methods of tension detection at a faster update rate. The system alleviates the problem of inadequate sampling rate using only machine vision methods and can be applied to future real-time control systems.

Intelligent transportation is a system that combines data-driven information with traffic management to achieve intelligent monitoring and retrieval functions. In order to further improve the retrieval accuracy of the system model, a new retrieval model was designed. The functional requirements of the system were summarized, and the three stages of data preprocessing, feature matching, and feature extraction were analyzed in detail. The study adopted preprocessing measures such as equalization and normalization to minimize the negative effects of noise and brightness. Based on the performance of various algorithms, the distance method was selected as the feature matching method, which has a wider applicability and is better at processing bulk data. Next, the study utilizes Euclidean distance method to extract keyframes and divides the feature extraction into three parts: color, shape, and texture. The methods of color moment, canny operator, and grayscale co-occurrence matrix are used to extract them, and ultimately achieve relevant image retrieval. The research conducted multiple experiments on the retrieval performance of the model, and analyzed the results of retrieving single and mixed features. The experimental results showed that the algorithm performed better in the face of mixed feature extraction. Compared with the average value of a single feature, the recall and precision of the three mixed features increased by 13.78% and 15.64%, respectively. Moreover, in the case of a large number of concurrent features, the algorithm also met the basic requirements. When the concurrent number was 100, the average response time of the algorithm is 4.46 seconds. Therefore, the algorithm proposed by the research institute effectively improves the ability of video retrieval and can meet the requirements of timeliness, which can be widely applied in practical applications.

The state-of-the-art capsule endoscopy (CE) technology offers painless examination for the patients and the ability to examine the interior of the gastrointestinal tract by a noninvasive procedure for the gastroenterologists. In this work, a modular and flexible CE development system platform consisting of a miniature field programmable gate array (FPGA) based electronic capsule, a microcontroller based portable data recorder unit and computer software is designed and developed. Due to the flexible and reprogrammable nature of the system, various image processing and compression algorithms can be tested in the design without requiring any hardware change. The designed capsule prototype supports various imaging modes including white light imaging (WLI) and narrow band imaging (NBI), and communicates with the data recorder in full duplex fashion, which enables configuring the image size and imaging mode in real time during examination. A low complexity image compressor based on a novel color-space is implemented inside the capsule to reduce the amount of RF transmission data. The data recorder contains graphical LCD for real time image viewing and SD cards for storing image data. Data can be uploaded to a computer or Smartphone by SD card, USB interface or by wireless Bluetooth link. Computer software is developed that decompresses and reconstructs images. The fabricated capsule PCBs have a diameter of 16mm. An ex-vivo animal testing has also been conducted to validate the results.

Feature point extraction and matching are the key modules in the front-end of visual simultaneous localization and mapping (SLAM), which determines the accuracy of locating and map Nbuilding robots. To obtain a good performance, the image extraction algorithm being applied should have high efficiency and robustness to the changes of the surrounding environment. In this paper, three widely used feature extraction algorithms, namely Scale-invariant feature transform (SIFT), speed up robust features (SURF) and oriented FAST, and rotated BRIEF (ORB), are studied. The basic concepts and details of the three algorithms are introduced. Then the execution time of the three algorithms is compared to evaluate their efficiency. Meanwhile, the key points matching rate of the three algorithms are compared under various transformations and distortions. The experiment results show that ORB has higher efficiency than SIFT and SURF. However, SIFT has better robustness to image rotation, illumination change, scaling transformation, Gaussian noise and salt-and-pepper noise. In addition, the performance of the ORB-SLAM system is tested on different datasets. It has great ability in camera trajectory estimation and map building and therefore is suitable for deploying in industry.

This paper concerns the problem of standing tree image mosaic based on SURF in pocket PC. In standing tree measurement based on image processing, the first step is to take photo of the tree to be measured. Due to some special tall trees, or limited by geographical conditions, it is impossible to get the full image of the tree for one time, and need to shoot sequence tree images with overlapped areas and stitch them. SIFT algorithm is good enough in image mosaic, but it is too complicated to work good in pocket PC. In this paper, a tree image mosaic method based on speed up robust features (SURF) was put forward embedded in pocket PC. The experiment of true tree image sequence stitching achieves the better smooth mosaic effect and shows the method is effective.

This book focuses on embedded image processing and, more specifically, on the efficient implementation of image processing algorithms on the TMS320C6000 family of digital signal processors (DSPs) from Texas Instruments (TI). The text is divided into six chapters: "Introduction," "Tools," "Spatial Processing Techniques," "Image Filtering," "Edge Detection and Segmentation," and "Wavelets." The text is supported by two appendices, as well as a CD-ROM. The CD-ROM includes all source code and project files as well as many test images and one text video clip. There is no better way to learn than by example, and the book offers plenty of them. While the book is intended for signal and image processing practitioners or software developers that plan to use a TI DSP, most of the programming techniques demonstrated in the book can easily be applied to other embedded platforms. The book is well written and succeeds in filling a big void in image processing literature, tackling how to efficiently implement signal and image processing algorithms using embedded processors.