The dramatic development of Edge Computing technologies is strongly stimulating the adoption of machine learning models on connected and autonomous vehicles (CAVs) so that they can provide a variety of intelligent onboard services. When multiple services running on the resource-constrained CAVs, how limited resources can dynamically support the desired services is of the utmost importance for both automakers and domain researchers. In this context, efficiently and dynamically managing vehicle services becomes critical for autonomous driving. While previous research focused on service scheduling, computation offloading, and virtual machine migration, we propose EdgeWare, an extensible and flexible middleware to manage the execution of vehicle services, which is open-source to the community with four key features: i) on-demand model switch, i.e., easily switch and upgrade machine learning models, ii) function consolidation and deduplication to eliminate duplicate copies of repeating functions and maximize the reusability of vehicle services, iii) build event-driven applications to reduce workload, and iv) dynamic workflow customization which enables customizing workflow to extend the functionality. Our experiment results show that EdgeWare accelerates the execution of services about 2.6 $$\times$$ faster compared to the silo approach and save CPU and memory utilization up to around 50% and 17% respectively, and it allows domain researchers to dynamically add new services on CAVs or easily switch to the upgraded applications for the life cycle management of vehicle services.

Geometric object detection has many applications, such as in tracking. Particle tracking microrheology is a technique for studying mechanical properties by accurately tracking the motion of the immersed particles undergoing Brownian motion. Since particles are carried along by these random undulations of the medium, they can move in and out of the microscope's depth of focus, which results in halos (lower intensity). Two-point particle tracking microrheology (TPM) uses a threshold to find those particles with peak, which leads to the broken trajectory of the particles. The halos of those particles which are out of focus are circles and the centres can be accurately tracked in most cases. When the particles are sparse, TPM will lose certain useful information. Thus, it may cause inaccurate microrheology. An efficient algorithm to detect the centre of those particles will increase the accuracy of the Brownian motion. In this paper, a hybrid approach is proposed which combines the steps of TPM for particles in focus with a circle detection step using circular Hough transform for particles with halos. As a consequence, it not only detects more particles in each frame but also dramatically extends the trajectories with satisfactory accuracy. Experiments over a video microscope data set of polystyrene spheres suspended in water undergoing Brownian motion confirmed the efficiency of the algorithm.

Since abnormal control chart patterns (CCPs) are indicators of production processes being out-of-control, it is a critical task to recognize these patterns effectively based on process measurements. Most methods on CCP recognition assume that the process data only suffers from single type of unnatural pattern. In reality, the observed process data could be the combination of several basic patterns, which leads to severe performance degradations in these methods. To address this problem, some independent component analysis (ICA) based schemes have been proposed. However, some limitations are observed in these algorithms, such as lacking of the capability of monitoring univariate processes with only one key measurement, misclassifications caused by the inherent permutation and scaling ambiguities, and inconsistent solution. This paper proposes a novel hybrid approach based on singular spectrum analysis (SSA) and support vector machine (SVM) to identify concurrent CCPs. In the proposed method, the observed data is first separated by SSA into multiple basic components, and then these separated components are classified by SVM for pattern recognition. The scheme is suitable for univariate concurrent CCPs identification, and the results are stable since it does not have shortcomings found in the ICA-based schemes. Furthermore, it has good generalization performance of dealing with the small samples. Superior performance of the proposed algorithm is achieved in simulations.

This paper introduces an algorithm to super-resolve an image based on a self-training filter (STF). As in other methods, we first increase the resolution by interpolation. The interpolated image has higher resolution, but is blurry because of the interpolation. Then, unlike other methods, we simply filter this interpolated image to recover some missing high frequency details by STF. The input image is first downsized at the same ratio used in super-resolution, then upsized. The super-resolution filters are obtained by minimizing the mean square error between the upsized image and the input image at different levels of the image pyramid. The best STF is chosen as the one with minimal error in the training phase. We have shown that STF is more effective than a generic unsharp mask filter. By combining interpolation and filtering, we achieved competitive results when compared to support vector regression methods and the kernel regression method.

Support vector regression has been proposed in a number of image processing tasks including blind image deconvolution, image denoising and single frame super-resolution. As for other machine learning methods, the training is slow. In this paper, we attempt to address this issue by reducing the feature dimensionality through Principal Component Analysis (PCA). Our single frame supper-resolution experiments show that PCA successfully reduces the feature dimensionality without degrading the performance of SVR when the training images and testing images share similarities (i.e. belong to the same category). In fact, in some cases the performance in terms of Peak Signal- to-Noise Ratio (PSNR), is even better.

Compression of training sets is a technique for reducing training set size without degrading classification accuracy. By reducing the size of a training set, training will be more efficient in addition to saving storage space. In this paper, an incremental clustering algorithm, the Leader algorithm, is used to reduce the size of a training set by effectively subsampling the training set. Experiments on several standard data sets using SVM and KNN as classifiers indicate that the proposed method is more efficient than CONDENSE in reducing the size of training set without degrading the classification accuracy. While the compression ratio for the CONDENSE method is fixed, the proposed method offers variable compression ratio through the cluster threshold value.

As many other inverse problems, single-frame image super-resolution is an ill-posed problem. The problem has been approached in the context of machine learning. However, the proposed method in this paper is different from other learning based methods regarding how the input/output are formulated as well as how the learning is done. The assumption behind example based methods is the local similarity across seemingly different images. The assumption is illustrated by examples of image coding. Because of the differences in formulating the input/output and the implementation of Support Vector Regression (SVR), it is shown that the proposed approach outperforms the competing SVR method and the kernel regression method in terms of Peak Signal-to-Noise Ratio (PSNR), objective measurements of image quality. Since example based approaches are based on training, in which we know exactly what the output shall be. Therefore, it is proper to objectively measure the performance since the trained model is expected to " correctly" restore the image rather than to enhance the image, e. g. sharpening.

Atmospheric turbulence is caused by the random fluctuations of the refraction index of the medium. It can lead to blurring in images acquired from a long distance away. Since the degradation is often not completely known, the problem is viewed as blind image deconvolution or blur identification. Our previous work has observed that blurring increases kurtosis and introduced a new blur identification method based on kurtosis minimization (KM). In this letter, this observation has been studied using phase correlation. The KM method is compared with two other signal processing methods. The limitation of the method is also discussed.

Dynamic geometric distortion has been observed in the video captured by optical sensors in the presence of atmospheric turbulence. The distortion is caused by gradients in refractive index in the atmosphere resulting from temperature gradients in the air. This problem has been addressed previously mainly through a reference video approach which cannot handle the cases where real motion exists. In this paper, we propose a new atmospheric turbulent motion suppression algorithm that is able to suppress the turbulent motion significantly while the natural motion of objects/camera such as panning/zooming in the video is preserved.

Support vector regression (SVR) has been applied for blind image deconvolution. In this correspondence, it is applied in the problem of image denoising. After training on noisy images with ground-truth, support vectors (SVs) are identified and their weights are computed. Then the SVs and their weights are used in denoising different images corrupted by random noise at different levels on a pixel-by-pixel basis. The proposed SVR based image denoising algorithm is an example-based approach since it uses SVs in denoising. The SVR denoising is compared with a multiple wavelet domain method (Besov ball projection). Some initial experiments indicate that SVR based image denoising outperforms Besov ball projection method on non-natural images (e.g. document images) in terms of both peak signal-to-noise ratio (PSNR) and visual inspection.

Motivated by the success of support vector regression (SVR) in blind image deconvolution, we apply SVR to single-frame super-resolution. Initial results show that even when trained on as little as a single image, SVR is able to learn a generally applicable model that can super-resolve dissimilar images.

In this paper, an example-based image denoising algorithm is introduced. Image denoising is formulated as a regression problem, which is then solved using support vector regression (SVR). Using noisy images as training sets, SVR models are developed. The models can then be used to denoise different images corrupted by random noise at different levels. Initial experiments show that SVR can achieve a higher peak signal-to-noise ratio (PSNR) than the multiple wavelet domain Besov ball projection method on document images.

Our earlier work revealed a connection between blind image deconvolution and principal components analysis (PCA). In this letter, we explicitly formulate multichannel and single-channel blind image deconvolution as a PCA problem. Although PCA is derived from blur models that do not contain additive noise, it can be justified on both theoretical and experimental grounds that the PCA-based restoration algorithm is actually robust to the presence of white noise. The algorithm is applied to the restoration of atmospheric turbulence-degraded imagery and compared to an adaptive Lucy-Richardson maximum-likelihood algorithm on both real and simulated atmospheric turbulence blurred images. It is shown that the PCA-based blind image deconvolution runs faster and is more robust to noise.

Atmospheric turbulence is a common problem in astronomy and long distance surveillance applications. It can lead to optical distortions that can significantly degrade the quality of the captured images and video. Quality improvement can be achieved through digital restoration methods that effectively suppress the effects of optical distortion. In this paper, atmospheric optical distortion is modeled as having two components: a dispersive component and a time-varying distortion component. A new restoration algorithm is introduced that compensates for dispersion using a fourth-order statistic and employs a new adaptive warping algorithm to suppress turbulent motion effects. The new algorithm is able to improve quality significantly and is able to handle diffcult cases involving panning, zooming, and natural motion.

Revision with unchanged content. The goal of the work is to compensate for the blurring and geometric distortions caused by atmospheric turbulence and to produce a clear restoration of the scene. Since the blurring is unknown, a blind image deconvolution technique was developed based on a higher order statistic. Geometric distortion is handled through motion trajectory smoothing built upon optical flows. The technique significantly improved the quality of atmospheric turbulence degraded video. It explicitly addressed the geometric distortion that is ignored in many previous works and it can handle real motion cases when the previous methods that do address this geometric distortion fail. The research advanced the study of restoration turbulence degraded video. The work is of considerable interest to atmospheric scientists, earth-based astronomers, and also for long-range surveillance. In astronomy and surveillance, this technique could help to overcome some of the degrading effects of turbulence. The research work in this topic has advantages over others currently used and proposed because it does not require the addition of expensive optical equipment and does not require reference objects.

The classification of documents with sparse text, and video analysis, relies on accurate image classification. We herein present a method for binary classification that accommodates any number of individual classifiers. Each individual classifier is defined by the critical point between its two means, and its relative weighting is inversely proportional to its expected error rate. Using 10 simple image analysis metrics, we distinguish a set of "natural" and "city" scenes, providing a "semantically meaningful" classification. The optimal combination of 5 of these 10 classifiers provides 85.8% accuracy on a small (120 image) feasibility corpus. When this feasibility corpus is then split into half training and half testing images, the mean accuracy of the optimum set of classifiers was 81.7%. Accuracy as high as 90% was obtained for the test set when training percentage was increased. These results demonstrate that an accurate classifier can be constructed from a large pool of simple classifiers through the use of the statistical ("Normal") classification method described herein.

In this paper, we describe an algorithm for identifying a parametrically described blur based on kurtosis minimization. Using different choices for the parameters of the blur, the noisy blurred image is restored using Wiener filter. We use the kurtosis as a measurement of the quality of the restored image. From the set of the candidate deblurred images, the one with the minimum kurtosis is selected. The proposed technique is tested in a simulated experiment on a variety of blurs including atmospheric turbulence blurs, Gaussian blurs, and out-of-focus blurs. The proposed approach is also tested on real blurred images. Moreover, we test the performance when a wrong blur model is given. Our experiments show that the kurtosis minimization measurements match well with methods that maximize PSNR.

The presence of optical turbulence in video acquired by cameras viewing scenes at long distances can contribute significantly to degradation. This problem arises routinely, for example, in astronomy where objects of interest reside beyond the earth's atmosphere. Optical turbulence introduces time-varying perturbations in the images as well as blurring. In this paper, we introduce a method for suppressing the effects of this turbulence to enhance the quality of the observed objects and scenes. The proposed method is based on a new form of adaptive control grid interpolation in which computed motion vectors are used as the basis for turbulence estimation and suppression. In particular, the quasi-periodicities of the turbulent motion are exploited in the algorithm, which allow them to be suppressed while true motion (such as panning and zooming) is preserved.