Action recognition in videos is one of the essential, challenging and active area of research in the field of computer vision that adopted in various applications including automated surveillance systems, security systems and human computer interaction. In this paper, we present an in-depth comparative analysis of five CNN-RNN models based on pre-trained networks such as InceptionV3, VGG16, MobileNetV2, ResNet152V2 and InceptionResNetV2 with recurrent LSTM units for action recognition on Anomaly-5 dataset. The performance of these models is analyzed and compared in terms of accuracy, precision, recall & F1-scores and computational efficiency. The CNN-RNN architectures we considered for analysis in this paper, the ResNet152V2 based CNN-RNN model exhibits better performance and achieved highest accuracy, precision, recall and F1-score equal to 92.20% due to its ability to capture more complex spatial features. This comparative analysis may guide the researchers in selecting appropriate models for real-world applications for action recognition. In addition of this, a new dataset is developed called Anomaly-5 that can helps as a valuable resource for training and evaluating action recognition algorithms.

Technological advancements in wireless communication have paved the way for remote monitoring of hospital patients using wearable biomedical sensors. The network of such sensor nodes is termed as Wireless Body Sensor Networks (WBSNs). The healthcare applications of WBSN are delay sensitive and require timely dissemination of patient related critical data. However, poor link quality may affect the Quality of Service (QoS), resulting in higher transmission delays that cause detrimental impact on the performance of such networks. Moreover, malicious interventions can breach privacy and confidentiality of patient data, thereby making reliable communication a challenging task. In addition, the wireless properties of in-vivo nodes cause electromagnetic radiations during transmission which are absorbed by human body, resulting in damage of sensitive tissues around the biosensor nodes. Considering these issues, we propose a reliable, link-quality and temperature-aware routing protocol, referred to as RLT, for WBSN. This multi-facet, QoS aware routing protocol routes the data through good quality links having reliable nodes en-route and avoids hotspot regions. The evaluation results presented in the paper demonstrate the efficacy of the RLT routing scheme against the state-of-the-art techniques.

The popularity of Product-Recommendation (PR) or system of recommendation is rising day by day. Product suggestions are an-ecommerce customization approach which goods are continuously created for a customer on a webpage, application, or email based on data such user characteristics, browsing behavior, or situational context, resulting in an individualized purchasing experience. The system of recommendation used to predict or recommend the product according to the taste of customer. In today's life product recommendation system has been used by different E-Commerce sites. A website that allows people to buy and sell physical things, services, and digital products without having to go to a physical store. Through an e-commerce website, a company can manage orders, payments, shipping and logistics, and customer service. Recommendation can be of any type such as for music recommendation there is Spotify, for movies Netflix, for videos YouTube, play store (for different categories) and so on. For the recommendation of product different filtering methods and algorithms were used, to recommend products according to user's likeness. In this paper discussed about the existing Machine Learning Techniques (MLT) which were used for the product recommendation. Through these techniques the algorithm is used to predict or similar items according to user's likeness based on his information.

Aggressive downscaling of MosFET devices has some limitation due to short channel effects. These short channel effects affect the electronic behavior of device adversely. FinFET was proposed to overcome this shortcoming of MosFETs. But like MosFETs, FinFET have a problem of variability, which becomes more and more effective with the scaling. Variability analysis was the main theme of this research. Oxide thickness variations were taken into account for this research. 5%, 10% and 25% of variations were applied to oxide thickness and results were obtained for drain current, leakage current and threshold voltage. Synopsys Sentaurus as well as in-house simulator (Green's function simulation) were used for variability analysis. (C) 2021 INT TRANS J ENG MANAG SCI TECH.

For the RF analog applications, the two variants of FinFETs i.e., Independent gate (IG) and Short-circuited gate (SG) must be analyzed to understand the effect of different biases at the two gates. The variability of the device due to process variations fluctuates in accordance to the FinFET bias, at least for DC output. This paper provides a novel and systematic study of variability focusing on AC parameters in both SG and IG conditions for a 2-dimensional FinFET. Green's Function technique which uses the linearization of the non-linear responses is used to conduct the research utilizing a quasilinear state for the study of nonlinear variability [1], [2]. Study of FinFET's AC variability involves physical and geometric parameters which are most relevant for our analysis, as the parasitics of the FinFET are significantly affected by these parameters. The parasitics also varied with the two variants of

A product recommendation system is aimed at providing an improved shopping experience to the users, thereby increasing the revenues. The text-based product search on most of the online stores is based on the past search history of the user and/or the other characteristics associated to the products, such as the annotated labels, price range, category, description, size, color, and other attributes. Though, this traditional method has adequate performance, however, it is prone to the issues pertaining to improper annotation of the metadata related to the products. Due to the recent advancements in machine learning, product recommendation using machine learning is getting increasing attention. Over the last one decade, tremendous progress in machine learning and computing has paved the way for efficient product recommendation. In this book chapter, we present a critical analysis of product recommendation using the traditional methods. Specifically, the book chapter explores the prospect of product recommendation based on visual similarity using machine learning which is rather a new concept in this domain and is getting increasing attention in research community. The chapter also explores the underlying challenges and discusses the future directives.

Due to the growing volume of multimedia data generated these days, it has become extremely difficult to manually analyze the data and extract useful information from it. Especially the analysis of videos pertaining to different fields such as surveillance, videos, social media, education, etc. cannot be done efficiently by manual methods. This requires automatic analysis algorithms that can intelligently analyze videos and derive salient information from them. This information can be useful in a number of tasks such as video segmentation, incident detection, anomaly detection, query-based video retrieval, and content censorship. This chapter provides a detailed review of the techniques proposed for video analysis to provide a compact set of video tags. This chapter considers it a joint tag-segmentation problem and critically analyzes the relevant literature to highlight their respective pros and cons. At the end, potential research areas in this domain and suggestions for improvement are discussed.

The sheer volume of movies generated these days requires an automated analytics for efficient classification, query-based search, and extraction of desired information. These tasks can only be efficiently performed by a machine learning based algorithm. We address the same issue in this paper by proposing a deep learning based technique for predicting the relevant tags for a movie and segmenting the movie with respect to the predicted tags. We construct a tag vocabulary and create the corresponding dataset in order to train a deep learning model. Subsequently, we propose an efficient shot detection algorithm to find the key frames in the movie. The extracted key frames are analyzed by the deep learning model to predict the top three tags for each frame. The tags are then assigned weighted scores and are filtered to generate a compact set of most relevant tags. This process also generates a corpus which is further used to segment a movie based on a selected tag. We present a rigorous analysis of the segmentation quality with respect to the number of tags selected for the segmentation. Our detailed experiments demonstrate that the proposed technique is not only efficacious in predicting the most relevant tags for a movie, but also in segmenting the movie with respect to the selected tags with a high accuracy.

Wireless body area networks (WBANs) are one of the applications of IoT that deal with the remote transmission of patient data. The health-related data is of highly sensitive nature. The loss of critical data packets might lead a patient to an embarrassing position. Therefore, WBANs require a secure and efficient data transmission mechanism. However, wireless transmission traditionally remains vulnerable to many security attacks such as node misbehavior attack. Routing protocols play a key role in the extended network lifetime. However, efficient data routing in WBAN is a challenging task. In addition, sensor nodes due to wireless communication produce electromagnetic radiation that is absorbed by the human body and results in temperature rise. Therefore, routing protocols along with security issues must consider temperature rise to ensure safe wireless transmission. In this chapter, the authors present a comprehensive review of most relevant security and privacy concerns and relevant routing protocols addressing the aforementioned routing issues.

This paper presents resource allocation algorithms for video transcoding service on a cloud. The main objective of the proposed algorithms is to allocate and de-allocate Virtual Machines (VMs) horizontally in a cluster of video transcoding servers. For cost-efficiency and better utilization of resources, video segmentation at group of pictures level is used. With video segmentation, a video is split into smaller segments that can be sent for transcoding on any transcoding server. To demonstrate the efficiency of the proposed algorithms, a discrete-event simulation is used. The proposed algorithms are also compared with the existing video transcoding algorithms in a cloud computing environment. The existing VM allocation algorithms are based on accumulative play rate and transcoding rate, while our improved proactive VM allocation algorithms also take into account the overall computation load and system throughput. The results indicate that the proposed algorithms are more cost-efficient than the existing algorithms.

Wireless Sensor Networks (WSN) have gained remarkable appreciations over the last few years. Despite significant advantages and tremendous applications, WSN is vulnerable to variety of attacks. Due to unattended nature of WSN, sensor nodes are more prone to be overtaken by an adversary. By doing that, an adversary can learn the contents of the victim’s memory, can have access to valid cryptographic keys, and can also modify the behavior of corrupted nodes. In this paper, we investigate some of the most severe node misbehavior attacks in WSN, namely blackhole and grayhole attacks, using Ad-hoc On Demand Distance Vector (AODV) routing protocol. A detailed NS2 based implementation and comparative analysis of these attacks has been presented. The performance of AODV is evaluated by considering different metrics such as packet delivery ratio, packet drop ratio, average end-to-end delay, normalized routing load, and energy consumption. Simulation results are provided to show the effects of these attacks on AODV protocol which suffers from increased packet loss and decreased delivery ratio. Some counter measures against node misbehavior attacks are also provided.

Routing protocols play a pivotal role in energy-efficient, reliable and robust communication in Mobile Ad hoc Networks (MANETs). In order to ensure efficient communication, the optimal operation setting of a routing protocol is essential to be ascertained. In this paper, we perform a comparative analysis of three most popular MANET routing protocols, namely, ad hoc on demand distance vector (AODV), dynamic source routing, and destination sequenced distance vector protocols. We evaluate the performance of these protocols for different network sizes, each with low and high traffic scenario. The generic evaluation criteria which specify the performance of routing protocols and used in our simulations include packet delivery ratio, end-to-end delay, average remaining energy of nodes, and throughput. Our in-depth analysis and the comparison results presented in this paper show that AODV protocol outperforms the other two protocols for the selected parameters and various network scenarios.

Dynamic power management (DPM) refers to strategies which selectively change the operational states of a device during runtime to reduce the power consumption based on the past usage pattern, the current workload, and the given performance constraint. The power management problem becomes more challenging when the workload exhibits nonstationary behavior which may degrade the performance of any single or static DPM policy. This article presents a reinforcement learning (RL)-based DPM technique for optimal selection of timeout values in the different device states. Each timeout period determines how long the device will remain in a particular state before the transition decision is taken. The timeout selection is based on workload estimates derived from a Multilayer Artificial Neural Network (ML-ANN) and an objective function given by weighted performance and power parameters. Our DPM approach is further able to adapt the power-performance weights online to meet user-specified power and performance constraints, respectively. We have completely implemented our DPM algorithm on our embedded traffic surveillance platform and performed long-term experiments using real traffic data to demonstrate the effectiveness of the DPM. Our results show that the proposed learning algorithm not only adequately explores the power-performance trade-off with nonstationary workload but can also successfully perform online adjustment of the trade-off parameter in order to meet the user-specified constraint.

In this paper, we investigate the implementation of image filtering in frequency domain using NVIDIA?s CUDA (Compute Unified Device Architecture). In contrast to signal and image filtering in spatial domain which uses convolution operations and hence is more compute-intensive for filters having larger spatial extent, the frequency domain filtering uses FFT (Fast Fourier Transform) which is much faster and significantly reduces the computational complexity of the filtering. We implement the frequency domain filtering on CPU and GPU respectively and analyze the speed-up obtained from the CUDA?s parallel processing paradigm. In order to demonstrate the efficiency of frequency domain filtering on CUDA, we implement three frequency domain filters, i.e., Butterworth, low-pass and Gaussian for processing different sizes of images on CPU and GPU respectively and perform the GPU vs. CPU benchmarks. The results presented in this paper show that the frequency domain filtering with CUDA achieves significant speed-up over the CPU processing in frequency domain with the same level of (output) image quality on both the processing architectures

This paper proposes a RL (Reinforcement Learning) based DPM (Dynamic Power Management) technique to learn timeout policies during a visual sensor node's operation which has multiple power/performance states. As opposed to the widely used static timeout policies, our proposed DPM policy which is also referred to as OLTP (Online Learning of Timeout Policies), learns to dynamically change the timeout decisions in the different node states including the non-operational states. The selection of timeout values in different power/performance states of a visual sensing platform is based on the workload estimates derived from a ML-ANN (Multi-Layer Artificial Neural Network) and an objective function given by weighted performance and power parameters. The DPM approach is also able to dynamically adjust the power-performance weights online to satisfy a given constraint of either power consumption or performance. Results show that the proposed learning algorithm explores the power-performance tradeoff with non-stationary workload and outperforms other DPM policies. It also performs the online adjustment of the tradeoff parameters in order to meet a user-specified constraint.

MobiTrick is a portable and compact traffic monitoring system that utilizes image processing capabilities to perform typical traffic monitoring tasks. It is based on a heterogeneous-sensors architecture using infrared and visible-light cameras. This setup allows utilizing the advantages of both sensors and additionally enables heterogeneous stereo reconstruction for 3D monitoring of vehicles. Due to the mobility factor, MobiTrick sensing platform is battery operated and hence imposes a strict limitation on power consumption. Therefore, it needs an efficient power management technique that optimizes the overall power consumption of the system. This paper presents MobiTrick’s design architecture, novel vision-based techniques to perform the monitoring tasks, and the current work on an online Dynamic Power Management (DPM) strategy to minimize the sensing platform’s power consumption.

Dynamic Power Management (DPM) refers to a set of strategies that achieves efficient power consumption by selectively turning off (or reducing the performance of) a system components when they are idle or are serving light workloads. This paper presents a Reinforcement Learning (RL) based DPM technique for a portable, multi-camera traffic monitoring system. We target the computing hardware of the sensing platform which is the major contributor to the entire power consumption. The RL technique used for the DPM of the sensing platform uses a model-free learning algorithm that does not require a priori model of the system. In addition, a robust workload estimator based on an online, Multi-Layer Artificial Neural Network (ML-ANN) is incorporated to the learning algorithm to provide partial information about the workload and to take better decisions according to the changing workload. Based on the estimated workload and a selected power-latency tradeoff parameter, the algorithm learns to use optimal time-out values in sleep and idle modes of the computing hardware. Our results show that the learning algorithm learns an optimal DPM policy for the non-stationary workload, while significantly reducing the power consumption and keeping the system response to a desired level.

The biggest challenges faced by intelligent traffic monitoring systems are mobility, compactness and energy-efficiency. Current traffic monitoring systems are based on fixed installations and thus have no or least portability. Also, they use many sensors (e.g, cameras, induction loops, radar or laser), utilize little or no image processing capabilities, and are difficult to set-up. As images contain a lot of information, the surveillance systems purely based on vision can help avoiding the use of additional sensors, reducing the size of the sensor platform and hence increasing the flexibility and mobility. Since mobile systems often run from batteries, power consumption is a major issue and these systems should be highly energy-efficient. In this paper, we describe the heterogeneous sensor architecture of our mobile traffic surveillance system MobiTrick and its potential dynamic power management. The use of heterogeneous sensors is motivated by utilizing the 3D stereo information from the heterogeneous visual sensors to perform the required operations and thus avoiding the use of other large sensors.

A benchmarking using cellular neural networks is performed between the traditional method of Genetic algorithm (using binary population of random chromosomes) with a real coded approach of genetic algorithm. The benchmarking was done with various image processing operations and it is shown that in most of the image processing operations, real coded appraoch converges faster. Real numbers population prevents the repeated encoding and decoding of chromosomes. Also the sizes of chromosomes are relatively smaller. Moreover, a modified type of 2-point crossover (F-Crossover) is introduced which decreases the convergence time of the genetic algorithm and eliminates the need of mutation.