Search

Simple Summary: The automatic diagnosis of fish diseases is an important task in smart aquaculture, and plays a key role in detecting fish conditions, understanding disease signs, and improving fish welfare and health. This paper reviews the latest advances in intelligent fish disease diagnosis techniques for aquatic applications. To acquire high-quality images to improve the accuracy of fish disease diagnosis, the paper reviews, in the second paragraph, how image-processing techniques can be used to acquire high-quality images. The third paragraph of this paper reviews the methods of intelligent fish disease diagnosis, including expert system detection, camera image detection, microscope image detection, spectral image detection, fluorescence image detection, ultrasonic image detection, and sensor detection methods. Additionally, the advantages and disadvantages of each detection method are summarized, including whether they cause damage to the fish body. Finally, we put forward the prospect of the intelligent detection of fish diseases and summarize the methods that may be used in the intelligent diagnosis of fish disease. Aquatic products, as essential sources of protein, have attracted considerable concern by producers and consumers. Precise fish disease prevention and treatment may provide not only healthy fish protein but also ecological and economic benefits. However, unlike intelligent two-dimensional diagnoses of plants and crops, one of the most serious challenges confronted in intelligent aquaculture diagnosis is its three-dimensional space. Expert systems have been applied to diagnose fish diseases in recent decades, allowing for restricted diagnosis of certain aquaculture. However, this method needs aquaculture professionals and specialists. In addition, diagnosis speed and efficiency are limited. Therefore, developing a new quick, automatic, and real-time diagnosis approach is very critical. The integration of image-processing and computer vision technology intelligently allows the diagnosis of fish diseases. This study comprehensively reviews image-processing technology and image-based fish disease detection methods, and analyzes the benefits and drawbacks of each diagnostic approach in different environments. Although it is widely acknowledged that there are many approaches for disease diagnosis and pathogen identification, some improvements in detection accuracy and speed are still needed. Constructing AR 3D images of fish diseases, standard and shared datasets, deep learning, and data fusion techniques will be helpful in improving the accuracy and speed of fish disease diagnosis. [ABSTRACT FROM AUTHOR]

Objectives: In the remote sensing (RS) big data era, intelligent interpretation of remote sensing images (RSI) is the key technology to mine the value of big RS data and promote several important applications. Traditional knowledge-driven RS interpretation methods, represented by expert systems, are highly interpretable, but generally show poor performance due to the interpretation knowledge being difficult to be completely and accurately expressed. With the development of deep learning in computer vision and other fields, it has gradually become the mainstream technology of RSI interpretation. However, the deep learning technique still has some fatal flaws in the RS field, such as poor interpretability and weak generalization ability. In order to overcome these problems, how to effectively combine knowledge inference and data learning has become an important research trend in the field of RS big data intelligent processing. Generally, knowledge inference relies on a strong domain knowledge base, but the research on RS knowledge graph (RS-KG) is very scarce and there is no available large-scale KG database for RSI interpretation now. Methods: To overcome the above considerations, this paper focuses on the construction and evolution of the RS-KG for RSI interpretation and establishes the RS-KG takes into account the RS imaging mechanism and geographic knowledge. Supported by KG in the RS field, this paper takes three typical RSI interpretation tasks, namely, zero-shot RSI scene classification, interpretable RSI semantic segmentation, and large-scale RSI scene graph generation, as examples, to discuss the performance of the novel generation RSI interpretation paradigm which couples KG and deep learning. Results and Conclusions: A large number of experimental results show that the combination of RS-KG inference and deep da‐ ta learning can effectively improve the performance of RSI interpretation. The introduction of RS-KG can effectively improve the interpretation accuracy, generalization ability, anti-interference ability, and interpretability of deep learning models. These advantages make RS-KG promising in the novel generation RSI interpretation paradigm. [ABSTRACT FROM AUTHOR]

Wildfire detection is a time-critical application since it can be challenging to identify the source of ignition in a short amount of time, which frequently causes the intensity of fire incidents to increase. The development of precise earlywarning applications has sparked significant interest in expert systems research due to this issue, and recent advances in deep learning for challenging visual interpretation tasks have created new study avenues. In recent years, the power of deep learning-based models sparked the researcher’s interests from a variety of fields. Specially, Convolutional Neural Networks (CNN) have become the most suited approach for computer vision tasks. As a result, in this paper we propose a CNN-based pipeline for classifying and verifying fire-related images. Our approach consists of two models, first of which classifies the input data and then the second model verifies the decision made by the first one by learning more robust representations obtained from a large masked auto encoder-based model. The verification step boosts the performance of the classifier with respect to false positives and false negatives. Based on extensive experiments, our approach proves to improve previous state-of-the-art algorithms by 3 to 4% in terms of accuracy. [ABSTRACT FROM AUTHOR]

To ensure the cutting speed during the cutting operation, this paper proposes a groove cutting speed inference planning system that relies on production experience and set parameters and is based on machine vision and a two-level fuzzy neural hybrid network. The overall structure of the inference system is designed, including the mechanical body, vision system, and fuzzy neural hybrid network. The contour information of the part is obtained using industrial cameras and digital image processing systems. The cutting speed of the trajectory segment is inferred based on the related processing parameters and the secondary fuzzy neural hybrid network. Finally, all of the processing parameters are transmitted to the PLC, so that the robot can work according to the predetermined displacement and speed. Simulations verify that the speed inference planning system offers certain advantages compared to the traditional one. The appearance of the speed inference planning realises independent design and planning of the cutting speed, and further ensures the unity of the cutting quality and cutting speed. This proposed method provides a new direction for the development and transformation of machining processes that rely on manual experience and in which expert systems cannot be used. [ABSTRACT FROM AUTHOR]

• A survey of kernel clustering methods is provided. • Inspiring from the mechanism of flight guidance in particle swarm a fast kernel clustering method. • Presenting a framework for an automatic parameter tuning for the proposed method. • Application of the proposed method to image and video segmentation tasks. • Developing enhancement strategies to improve image and video segmentation results. Machine learning has become a pillar of today's expert and intelligent systems where a special attention has been drawn to unsupervised methods. Support vector data description is one of the most interesting methods proposed in this context that addresses two main challenges: the absence of ground truth labels and the presence of outliers in data. It is well known that the ultimate problem has a great impact on a system performance since it typically prevents and limits its generalization and hence its failure. Even though the strengths of the Support vector data description method, its application into multi class and large scale context, as computer vision systems, raises bottleneck issues due to its runtime complexity and mono-class context application. We propose, in this paper, an approach inspired from the movement behavior of particle swarm and relies on the use of core set paradigm to overcome the aforementioned drawbacks of Support vector data description method. Indeed, since the proposed method uses some parameters corresponding to kernel function and swarm's movement formulation, we present and study in this paper some strategies to ensure a dynamic adaptation of these parameters. The main qualities of the proposed method is that it provides a fast scheme for clustering a set of points without neither a prior knowledge on the number of naturally occurring cluster in the data nor an assumption on clusters shapes. A quantitative and comparative performance assessment is carried out over the Berkeley and BuffaloXiph datasets. The results witness the efficiency of the method in terms of both classification accuracy and computational performance. These quantitative assessments reinforce the significance as well as the importance of embedding the proposed method in other intelligent systems application areas. [ABSTRACT FROM AUTHOR]

Highlights • A new multi-agent dynamic system MADS architecture for the complex computer vision tasks. • An adaption of the proposed architecture to online multi-face tracking in video sequences. • A logical representation model of relations and mental attitudes of agents involved in the MADS. • The integration of two cognitively different type of agents for specific problem solving. • The integration of state-of-the-art methods for face detection and tracking with BDI agents. Abstract The paper presents a new architecture framework in the field of expert and intelligent systems which is based on four paradigms: a novel multi-agent dynamic system architecture (MADS), an extended Belief Desire Intention (EBDI) agent community, autonomy-oriented entities (AOEs), and deep learning concepts. The main impact of the proposed framework is a new approach, or even a new way of thinking, which enables integration of the concepts of deep learning, a conventional approach to solving the domain problem, cognitive agents with mental attitudes, and concepts of nature-inspired computing. All these allow the effective use of the framework in the field of intelligent and expert systems. The significance of the framework lies in its flexibility and adaptability based on the formal logical description of EBDI agents, the definition of the behaviour of AOEs, the use of classical modules for domain problem solving, and modules based on deep-learning concepts. We believe that the example of the adaptation of the proposed architecture framework to robust multi-face tracking illustrates the significance of the proposed framework. In this paper, MADS is adapted to the first two stages of a face de-identification pipeline: robust face detection and multi-face tracking. The proposed architecture of MADS has a two-level hierarchical organization. At the first level there is a manager designed as an Extended Belief Desire Intention (mEBDI) agent. The extension of a manager BDI agent consists of a convolutional neural network-based face detector, a set of autonomous-oriented entities for the elimination of false positive face detections, and a trajectory memory. At the second level, there are many tracking agents (trEBDIs) which consist of a basic BDI agent extended with a face tracker based on position and scale correlation filters, a visual appearance memory, and a trajectory memory. The mEBDI and trEBDI agents are defined by the modal logic and are described at the implementation level. The proposed architecture for a robust multi-face tracking system was tested on a subset of YouTube music videos. The qualitative results, as well as the preliminary quantitative results expressed by the standard testing metrics, demonstrate the effective adaptation of the proposed multi-agent dynamic architecture to a robust multi-face tracking system. [ABSTRACT FROM AUTHOR]

In the recent years, the computer vision community has shown great interest on depth-based applications thanks to the performance and flexibility of the new generation of RGB-D imagery. In this paper, we present an efficient background subtraction algorithm based on the fusion of multiple region-based classifiers that processes depth and color data provided by RGB-D cameras. Foreground objects are detected by combining a region-based foreground prediction (based on depth data) with different background models (based on a Mixture of Gaussian algorithm) providing color and depth descriptions of the scene at pixel and region level. The information given by these modules is fused in a mixture of experts fashion to improve the foreground detection accuracy. The main contributions of the paper are the region-based models of both background and foreground, built from the depth and color data. The obtained results using different database sequences demonstrate that the proposed approach leads to a higher detection accuracy with respect to existing state-of-the-art techniques. [ABSTRACT FROM AUTHOR]

Computer vision applications require choosing operators and their parameters, in order to provide the best outcomes. Often, the users quarry on expert knowledge and must experiment many combinations to find manually the best one. As performance, time and accuracy are important, it is necessary to automate parameter optimization at least for crucial operators. In this paper, a novel approach based on an adaptive discrete cuckoo search algorithm (ADCS) is proposed. It automates the process of algorithms' setting and provides optimal parameters for vision applications. This work reconsiders a discretization problem to adapt the cuckoo search algorithm and presents the procedure of parameter optimization. Some experiments on real examples and comparisons to other metaheuristic-based approaches: particle swarm optimization (PSO), reinforcement learning (RL) and ant colony optimization (ACO) show the efficiency of this novel method. [ABSTRACT FROM AUTHOR]

Yoga is a traditional Indian exercise. It specifies various body postures called asanas , practicing them is beneficial for the physical, mental, and spiritual well-being. To support the yoga practitioners, there is a need of an expert yoga asanas recognition system that can automatically analyze practitioner's postures and could provide suitable posture correction instructions. This paper proposes YogNet, a multi-person yoga expert system for 20 asanas using a two-stream deep spatiotemporal neural network architecture. The first stream utilizes a keypoint detection approach to detect the practitioner's pose, followed by the formation of bounding boxes across the subject. The model then applies time distributed convolutional neural networks (CNNs) to extract frame-wise postural features, followed by regularized long short-term memory (LSTM) networks to give temporal predictions. The second stream utilizes 3D-CNNs for spatiotemporal feature extraction from RGB videos. Finally, the scores of two streams are fused using multiple fusion techniques. A yoga asana recognition database (YAR) containing 1206 videos is collected using a single 2D web camera for 367 min with the help of 16 participants and contains four view variations i. e. front, back, left, and right sides. The proposed system is novel as this is the earliest two-stream deep learning-based system that can perform multi-person yoga asanas recognition and correction in realtime. Simulation result reveals that YogNet system achieved 77.29%, 89.29%, and 96.31% accuracies using pose stream, RGB stream, and via fusion of both streams, respectively. These results are impressive and sufficiently high for recommendation towards general adaption of the system. [ABSTRACT FROM AUTHOR]

Scene object recognition is an essential requirement for intelligent mobile robots. In addition to geometric or appearance features, modern recognition systems strive to incorporate contextual information , normally modelled through Probabilistic Graphical Models (PGMs) or Semantic Knowledge (SK). However, these approaches, separately, show some weaknesses that limit their application, e.g., the exponential complexity of the probabilistic inference over PGMs or the inability of SK to handle uncertainty. This paper presents a hybrid PGM-SK system for object recognition that integrates both techniques reducing their individual limitations and gaining in probabilistic inference efficiency, performance robustness, uncertainty handling, and providing coherent results according to domain knowledge codified by a human expert. We support this claim with an extensive experimental evaluation according to both recognition success and time requirements in real scenarios from two datasets (NYU2 and UMA-offices). The yielded figures support the suitability of the hybrid PGM-SK recognition system, and its applicability to mobile robotic agents. [ABSTRACT FROM AUTHOR]

Computer vision is an interdisciplinary field that includes methods for acquiring, processing, analyzing, and understanding visual information. In computer vision, usually the approaches to solve problems are specific-application methods and, therefore, reusing captured knowledge in computer vision is hard. However, the aim of knowledge modeling (KM) is to capture and reuse knowledge to solve different problems. In this paper, we propose a knowledge-based component library for computer vision tasks such as video surveillance applications. The developed components are based on the region of interest (ROI) a well-known concept in the image processing and computer vision fields. We provide a set of reusable components that are specializations and/or compositions of ROIs. Finally, we propose several case studies that illustrate the feasibility of the proposal. Experimental results show that the proposed method deals effectively and efficiently with real-life computer vision problems. [ABSTRACT FROM AUTHOR]

In recent years there has been a sharp rise in applications, in which significant events need to be classified but only a few training instances are available. These are known as cases of one-shot learning. To handle this challenging task, organizations often use human analysts to classify events under high uncertainty. Existing algorithms use a threshold-based mechanism to decide whether to classify an object automatically or send it to an analyst for deeper inspection. However, this approach leads to a significant waste of resources since it does not take the practical temporal constraints of system resources into account. By contrast, the focus in this paper is on rigorously optimizing the resource consumption in the system which applies to broad application domains, and is of a significant interest for academic research, industrial developments, as well as society and citizens benefit. The contribution of this paper is threefold. First, a novel Deep Reinforcement One-shot Learning (DeROL) framework is developed to address this challenge. The basic idea of the DeROL algorithm is to train a deep-Q network to obtain a policy which is oblivious to the unseen classes in the testing data. Then, in real-time, DeROL maps the current state of the one-shot learning process to operational actions based on the trained deep-Q network, to maximize the objective function. Second, the first open-source software for practical artificially intelligent one-shot classification systems with limited resources is developed for the benefit of researchers and developers in related fields. Third, an extensive experimental study is presented using the OMNIGLOT dataset for computer vision tasks, the UNSW-NB15 dataset for intrusion detection tasks, and the Cleveland Heart Disease Dataset for medical monitoring tasks that demonstrates the versatility and efficiency of the DeROL framework. [ABSTRACT FROM AUTHOR]

Abstract: This paper proposes an automatic expert system for accuracy crop row detection in maize fields based on images acquired from a vision system. Different applications in maize, particularly those based on site specific treatments, require the identification of the crop rows. The vision system is designed with a defined geometry and installed onboard a mobile agricultural vehicle, i.e. submitted to vibrations, gyros or uncontrolled movements. Crop rows can be estimated by applying geometrical parameters under image perspective projection. Because of the above undesired effects, most often, the estimation results inaccurate as compared to the real crop rows. The proposed expert system exploits the human knowledge which is mapped into two modules based on image processing techniques. The first one is intended for separating green plants (crops and weeds) from the rest (soil, stones and others). The second one is based on the system geometry where the expected crop lines are mapped onto the image and then a correction is applied through the well-tested and robust Theil–Sen estimator in order to adjust them to the real ones. Its performance is favorably compared against the classical Pearson product–moment correlation coefficient. [Copyright &y& Elsevier]

Abstract: This paper is about the development of an expert system for automatic classification of granite tiles through computer vision. We discuss issues and possible solutions related to image acquisition, robustness against noise factors, extraction of visual features and classification, with particular focus on the last two. In the experiments we compare the performance of different visual features and classifiers over a set of 12 granite classes. The results show that classification based on colour and texture is highly effective and outperforms previous methods based on textural features alone. As for the classifiers, Support Vector Machines show to be superior to the others, provided that the governing parameters are tuned properly. [Copyright &y& Elsevier]

Abstract: Automatic container-code recognition is of great importance to the modern container management system. Similar techniques have been proposed for vehicle license plate recognition in past decades. Compared with license plate recognition, automatic container-code recognition faces more challenges due to the severity of nonuniform illumination and invalidation of color information. In this paper, a computer vision based container-code recognition technique is proposed. The system consists of three function modules, namely location, isolation, and character recognition. In location module, we propose a text-line region location algorithm, which takes into account the characteristics of single character as well as the spatial relationship between successive characters. This module locates the text-line regions by using a horizontal high-pass filter and scanline analysis. To resolve nonuniform illumination, a two-step procedure is applied to segment container-code characters, and a projection process is adopted to isolate characters in the isolation module. In character recognition module, the character recognition is achieved by classifying the extracted features, which represent the character image, with trained support vector machines (SVMs). The experimental results demonstrate the efficiency and effectiveness of the proposed technique for practical usage. [Copyright &y& Elsevier]

Abstract: Texture recognition have received tremendous attentions in the past decades, due to its wide applications in computer vision and pattern recognition. For various applications, formulating texture features in distributional forms can sometimes provide meaningful representation than in numerical forms. In this paper, a generalized probabilistic decision-based neural network (GPDNN), based on a novel methodology for the measurement of the difference between two distributions, is proposed for texture recognition. Based on a two-layer pyramid-type network structure, the proposed GPDNN receives texture data via 2-D grid input nodes, and outputs the classification and/or retrieval results at the top layer node. Our prototype system demonstrates a successful utilization of GPDNN to the texture recognition on 40 texture images selected from the MIT Vision Texture (VisTex) database. Regarding the performance, experiment results show that (1) based on the proposed distribution difference measurement method, the texture retrieval accuracy is improved from 77% to 82% by comparing with some recently published leading methods, and (2) the proposed GPDNN has significant improvements in classification accuracy from 82.2% to 90.1% and retrieval accuracy from 79.9% to 88.6% by comparing with traditional approaches. [ABSTRACT FROM AUTHOR]

Abstract: This paper focuses on the combination of variants of local binary patterns (LBP), widely considered the state of the art among texture descriptors, using the same radius and the same number of neighborhoods. We report new experiments exploring several LBP-based descriptors and propose a set of variants for the representation of images. Our experiments are of two main types. In the first set, the Fourier transform is used to extract features starting from the histogram of uniform patterns. In these experiments we test different methods of extracting features from the histogram and each method is used to train a set of support vector machines (SVMs) which are then combined. In the second set of experiments, features are extracted from histograms using different definitions of uniform patterns. These are used to train SVMs, and the results are then combined. Our results show that descriptors extracted from LPB using the same radius and the same number of neighborhoods can be combined to improve classifier performance. [ABSTRACT FROM AUTHOR]

Many algorithms for applications such as pattern recognition, computer vision, and computer graphics seek to compute actual optimal paths in weighted directed graphs. The standard approach for reporting an actual optimal path is based on building a single-source optimal path tree. A technique by Chen et al.2 was given for a class of problems such that a single actual optimal path can be reported without maintaining any single-source optimal path tree, thus significantly reducing the space bound of those problems with no or little increase in their running time. In this paper, we extend the technique by Chen et al.2 to the generalized problem of reporting many actual optimal paths with different starting and ending vertices in certain directed graphs, and show how this new technique yields improved results on several application problems, such as reconstructing a 3-D surface band bounded by two simple closed curves, finding various constrained segmentation of 2-D medical images, and circular string-to-string correction. We also correct an error in the time/space complexity for the well-cited circular string-to-string correction algorithm12 and give an improved result for this problem. Although the generalized many-path problem seems more difficult than the single-path cases, our algorithms have nearly the same space and time bounds as those of the single-path cases. Our technique is likely to help improve many other optimal paths or dynamic programming algorithms. [ABSTRACT FROM AUTHOR]

Tracking algorithms in computer vision usually fail when dealing with complex scenarios. This paper presents an extension of a general tracking system that uses context knowledge to solve tracking issues. The context layer represents knowledge about the context of the analyzed scenario and applies rules to reason with it, in order to assess the general tracking layer at different stages and enhance tracking results. The context knowledge representation and the reasoning methods are general and can be easily adapted to different scenarios. The experimentation results show that the performance of the tracking system is considerably improved, while the efficiency requirements that are mandatory in real-time systems are satisfied. [Copyright &y& Elsevier]

Abstract: This paper introduces a new vision-based and web-based mobile robot platform. The platform consists of control and communication centers, a mobile robot and real-time support libraries. All activities in the platform are achieved by only computer vision techniques. The platform provides monitoring, tele-controlling and programming for real-time educational exercises and helps to the users to achieve these exercises through a standard web browser without any need for additional support software. The results have shown that the proposed, designed and implemented platform provide amazing new facilities and features to the users (students and researchers) in applying their real-time exercises on web. [Copyright &y& Elsevier]

Abstract: This paper describes the knowledge modelling for a complex object tracking system based on a video sequence. By complex object tracking problems, we mean those based on articulated models and also the possibility of tracking multiple targets. The general proposed framework to solve the considered problem, referred as the “Visual Tracking” task, is based on a synergic combination of strategies coming from particle filters and population-based metaheuristics. The considered system works with a population of individuals (solutions) that evolve along time. These individuals cooperate among them and also improve their respective fitness values in order to offer an efficient near-optimal solution for each frame in the tracked video sequence. The three main resulting subtasks, namely “Extract”, “Explore” and “Exploit‘”, are described in detail by CommonKADS library schemes. We argue that the knowledge modelling components used in the decomposition of the Visual Tracking task can be reused as generic elements of problem-solving methods in video processing. [Copyright &y& Elsevier]

Natural image prior is one of the most efficient ways to represent images for computer vision tasks. In the literature, filter response statistics prior and synthesis-based sparse representation are two dominant prior models, which have been investigated separately and our knowledge of the relation between these two methods remains limited. In this paper, we examine the inherent relationship between the Fields of Experts (FoE) and K-SVD methods in the pursuit of natural image priors. We theoretically analyze and show that these two prior models have a mutually complementary relationship in the pursuit of the structure of natural images space. Based on these findings, a novel joint statistical prior is proposed, in which adaptive filters are obtained by exploring clues from both priors and utilized to characterize the subtle structure of natural images subspace. Qualitative and quantitative experiments demonstrate that the proposed method achieves a more comprehensive and reliable estimation of natural image prior and is competitive to both alternative and state-of-the-art methods. • Inherent relationship between the FoE and K-SVD methods for natural image priors is examined. • A novel joint statistical prior is proposed with adaptive filters. • Reliable estimation of natural image prior is achieved over the state-of-the-arts. [ABSTRACT FROM AUTHOR]

• Image-to-image transformation network is proposed for face and scene datasets. • The cyclic-synthesized loss is computed between the synthesized and cycled image. • The cyclic-synthesized loss enforces constraint to generate more realistic images. • Four publicly available paired datasets and five state-of-the art methods are used. • Superior results are observed as compared to the existing methods. The primary motivation of image-to-image transformation is to convert an image of one domain to another domain. The Generative Adversarial Network (GAN) is the recent trend for image-to-image transformation. The existing GAN models suffer due to the lack of utilization of proper synthesization objectives. In this paper, we propose a new Cyclic-Synthesized Generative Adversarial Networks (CSGAN) for the development of expert and intelligent systems for image-to-image transformation. The proposed CSGAN uses a new objective function based on the proposed cyclic-synthesized loss between the synthesized image of one domain and cycled image of another domain. The proposed CSGAN enforces the mapping from one domain to another domain more accurately by limiting the scope of redundant transformation with the help of the cyclic-synthesized loss. The performance of the proposed CSGAN is evaluated on four benchmark image-to-image transformation datasets, including CUHK Face dataset, WHU-IIP Thermal-Visible Face Dataset, CMP Facades dataset, and NYU-Depth Dataset. The results are computed using the widely used evaluation metrics such as MSE, SSIM, PSNR, and LPIPS. The experimental results of the proposed CSGAN approach are compared with the latest state-of-the-art approaches, such as GAN, Pix2Pix, DualGAN, CycleGAN, and PS2GAN. The proposed CSGAN technique outperforms all the methods over CUHK dataset, WHU-IIP dataset, NYU-Depth dataset, and exhibits promising and comparable performance over Facades dataset in terms of both qualitative and quantitative measures. The code is available at https://github.com/KishanKancharagunta/CSGAN. [ABSTRACT FROM AUTHOR]

Traffic light detection and recognition (TLR) research has grown every year. In addition, Machine Learning (ML) has been largely used not only in traffic light research but in every field where it is useful and possible to generalize data and automatize human behavior. ML algorithms require a large amount of data to work properly and, thus, a lot of computational power is required to analyze the data. We argue that expert knowledge should be used to decrease the burden of collecting a huge amount of data for ML tasks. In this paper, we show how such kind of knowledge was used to reduce the amount of data and improve the accuracy rate for traffic light detection and recognition. Results show an improvement in the accuracy rate around 15%. The paper also proposes a TLR device prototype using both camera and processing unit of a smartphone which can be used as a driver assistance. To validate such layout prototype, a dataset was built and used to test an ML model based on adaptive background suppression filter (AdaBSF) and Support Vector Machines (SVMs). Results show 100% precision rate and recall of 65%. [ABSTRACT FROM AUTHOR]

• Physics based model for crowd flow analysis. • Detection of linear motion flows using Langevin Equation-based model. • Better segmentation accuracy than existing Crowd Flow Segmentation methods. • Low Computational Overhead. Expert and intelligent systems are highly popular for designing cross-domain autonomous systems. Computer vision aided by expert decision-making systems are widely used to automate tasks that are normally carried out manually. For example, in the traditional way of traffic monitoring, signals are either controlled through predefined set-ups or with the help of visual observations. Even though some of the modern cities employ sensor-based surveillance at large, full automation is still far from the desirable accuracy. This is more challenging for monitoring high-density crowds that often cause unwanted situations due to sudden changes in dynamics. It has been shown in this paper that the movement of a dense crowd can be approximated using well-known physics-based models. Such a modeling can help to understand the overall crowd behavior. In accomplishing this, we have introduced a computer vision guided expert system with the help of a Langevin equation-based force model to represent the linear flow of the crowd, particularly in situations when the density is high. One of the primary contributions of our proposed model is its computational efficiency, particularly when a timely decision can help to avoid unwanted situations. Our proposed three-term force model is capable of predicting the positions of the group of key-points in a video frame leading to a significant computational gain. We have carried out several experiments on publicly available videos as well as our own videos to validate the claims in terms of accuracy as well as computational gain. It has been observed that the proposed physics-based model outperforms the existing systems with a 4 − 6 % improvement in the segmentation accuracy. Moreover, we have achieved multi-fold computational gain. We believe the proposed work, when supported by appropriate post-processing, can be used to develop crowd monitoring applications. [ABSTRACT FROM AUTHOR]

• Introduce a real-time expert system for anomaly detection of aerators. • Propose a novel RF-KLT algorithm for motion feature extraction in fixed region. • Develop a small object region detection method in complex background. • Present a time series dimension reduction method to build feature dataset. • Object region and working state detection accuracy is 100% and 99.9% respectively. Aerators are essential and crucial auxiliary devices in intensive culture, especially in industrial culture in China. In this paper, we propose a real-time expert system for anomaly detection of aerators based on computer vision technology and existing surveillance cameras. The expert system includes two modules, i.e., object region detection and working state detection. First, we present a small object region detection method based on the region proposal idea. Moreover, we propose a novel algorithm called reference frame Kanade-Lucas-Tomasi (RF-KLT) algorithm for motion feature extraction in fixed regions. Then, we describe a dimension reduction method of time series for establishing a feature dataset with obvious boundaries between classes. Finally, we use machine learning algorithms to build the feature classifier. The proposed expert system can realize real-time, robust and cost-free anomaly detection of aerators in both the actual video dataset and the augmented video dataset. Demo is available at https://youtu.be/xThHRwu_cnI. [ABSTRACT FROM AUTHOR]

• A new taxonomy of existing cosegmentation methods according to their frameworks. • For each method, we present adopted features and model, advantages and limitations. • A numerical comparison of relevant existing cosegmentation methods is proposed. • The performance of various methods was evaluated on five challenging datasets. • iCoseg, MSRC, Internet, FlickrMFC and PASCAL-VOC datasets are used for evaluation. Segmenting the foreground objects from an image is an essential low-level step for many expert and intelligent systems, and the success of this key process largely depends on the amount of available training data. However, the cost for obtaining annotations is still a bottleneck, requiring enormous human effort. Thus, in order to obtain higher segmentation accuracy without the need of annotations, computer vision researchers attempted to segment simultaneously the common regions from collections of images. Indeed, when images that share common foreground objects are available, we talk about cosegmentation. The task of cosegmentation can be considered as weakly supervised segmentation, seen that the input image set is sharing common foreground objects. In fact, several methods have been proposed while displaying different frameworks, models and strategies of work, for enhancing the cosegmentation accuracy. This work presents a comprehensive review of relevant cosegmentation methods, carried out extensively and in depth and completed by performance measurements. Indeed, the existing cosegmentation methods are roughly classified in this paper into the following categories: Markov random fields-based cosegmentation, Co-saliency-based cosegmentation, Image decomposition-based cosegmentation, Random Walker-based cosegmentation, Maps-based cosegmentation, Active contours-based cosegmentation, Clustering-based cosegmentation and Deep learning-based cosegmentation. For each category, we discuss the most relevant methods while presenting the proposed problem to deal, the degree of supervision, the adopted features, the number and the complexity of input images as well as the presented models and frameworks. Then, in order to exhibit and evaluate objectively and comprehensively the most relevant state-of-the-art methods, we provide a comparative study on various challenging datasets (iCoseg, MSRC, Internet, FlickrMFC and PASCAL-VOC datasets). The objective is to show the performances and the limitations of these methods, while using different metrics and providing the computation cost for many methods. In addition, we discuss various cosegmentation challenges, issues and some applications, what can be useful and helpful to understand the cosegmentation problem. Finally, we suggest some research directions for future research on image cosegmentation. [ABSTRACT FROM AUTHOR]

• We propose a multi-dimensional deep learning-based method for gesture recognition. • Learning spatiotemporal features from RGB-D videos and their motion representation. • We investigate different fusion strategies to boost the recognition performance. • We show efficiency to other video classification tasks (i.e. activity recognition). • The proposed method achieves the state-of-the-art results on several datasets. Human gesture recognition has become a pillar of today's intelligent Human-Computer Interfaces as it typically provides more comfortable and ubiquitous interaction. Such expert system has a promising prospect in various applications, including smart houses, gaming, healthcare, and robotics. However, recognizing human gestures in videos is one of the most challenging topics in computer vision, because of some irrelevant environmental factors like complex background, occlusion, lighting conditions, and so on. With the recent development of deep learning, many researchers have addressed this problem by building single deep networks to learn spatiotemporal features from video data. However, the performance is still unsatisfactory due to the limitation that the single deep networks are incapable of handling these challenges simultaneously. Hence, the extracted features cannot efficiently capture both relevant shape information and detailed spatiotemporal variation of the gestures. One solution to overcome the aforementioned drawbacks is to fuse multiple features from different models learned on multiple vision cues. Aiming at this objective, we present in this paper an effective multi-dimensional feature learning approach, termed as MultiD-CNN, for human gesture recognition in RGB-D videos. The key to our design is to learn high-level gesture representations by taking advantages from Convolutional Residual Networks (ResNets) for training extremely deep models and Convolutional Long Short-Term Memory Networks (ConvLSTM) for dealing with time-series connections. More specifically, we first construct an architecture to simultaneously learn the spatiotemporal features from RGB and depth sequences through 3D ResNets which are then linked to a ConvLSTM to capture the temporal dependencies between them, and we show that they better combine appearance and motion information effectively. Second, to alleviate distractions from background and other variations, we propose a method that encodes the temporal information into a motion representation, while a two-stream architecture based on 2D-ResNets is then employed to extract deep features from this representation. Third, we investigate different fusion strategies at different levels for blending the classification results, and we show that integrating multiple ways of encoding the spatial and temporal information leads to a robust and stable spatiotemporal feature learning with better generalization capability. Finally, we perform different experiments to evaluate the performance of the investigated architectures on four kinds of challenging datasets, demonstrating that our approach is particularly impressive where it outperforms prior arts in both accuracy and efficiency. The obtained results affirm also the importance of embedding the proposed approach in other intelligent systems application areas. [ABSTRACT FROM AUTHOR]