Browsing by Author "Guddeti, R.M.R."

Now showing 1 - 20 of 49

A framework for low cost, ubiquitous and interactive smart refrigerator
(Springer, 2024) Mundody, S.; Guddeti, R.M.R.
Internet of Things (IoT) and Artificial Intelligence (AI)-enabled technologies are essential in developing innovative environments and intelligent applications. IoT and AI-enabled appliances are entering our kitchens, adding more comfort and usability. However, these appliances are not economical and are beyond the reach of a commoner with a moderate income. An intelligent fridge is one such appliance. This paper proposes a design for developing a cost-effective, ubiquitous, and intelligent refrigerator. Unlike existing approaches, the proposed method identifies and predicts the fridge items based on Night Vision images and provides minimal natural language interaction with the fridge. The proposed design aims to convert any standard refrigerator into its more intelligent counterpart with minimal hardware and software requirements. The design allows users to view fridge contents on the go using a mobile application and interact with it using natural language. The transfer learning technique enables us to use a YOLOv5n model for object detection. As there are no publicly available Night Vision image datasets of fridge items, we created a custom dataset of Night Vision images to train and validate the object recognition model. Our model for object detection achieved a mAP of 97.1% compared to the YOLOv3-tiny and YOLOv4-tiny models, whose mAP are 94.8% and 96.3%, respectively. The overall cost of the refrigerator after deployment of the module is less than $300, making it an affordable option. The proposed framework meets most of the requirements of a low-cost, ubiquitous, interactive smart refrigerator. © 2023, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.
A Hybrid Bio-Inspired Algorithm for Scheduling and Resource Management in Cloud Environment
(Institute of Electrical and Electronics Engineers, 2020) Domanal, S.G.; Guddeti, R.M.R.; Buyya, R.
In this paper, we propose a novel HYBRID Bio-Inspired algorithm for task scheduling and resource management, since it plays an important role in the cloud computing environment. Conventional scheduling algorithms such as Round Robin, First Come First Serve, Ant Colony Optimization etc. have been widely used in many cloud computing systems. Cloud receives clients tasks in a rapid rate and allocation of resources to these tasks should be handled in an intelligent manner. In this proposed work, we allocate the tasks to the virtual machines in an efficient manner using Modified Particle Swarm Optimization algorithm and then allocation / management of resources (CPU and Memory), as demanded by the tasks, is handled by proposed HYBRID Bio-Inspired algorithm (Modified PSO + Modified CSO). Experimental results demonstrate that our proposed HYBRID algorithm outperforms peer research and benchmark algorithms (ACO, MPSO, CSO, RR and Exact algorithm based on branch-and-bound technique) in terms of efficient utilization of the cloud resources, improved reliability and reduced average response time. © 2008-2012 IEEE.
A Hybrid Fog-Cloud Framework for Smart Refrigerator Inventory Management
(Institute of Electrical and Electronics Engineers Inc., 2024) Mundody, S.; Guddeti, R.M.R.
Technological advances have led to the proliferation of smart home applications, which rely on sensors. The sensors used in smart home applications generate vast amounts of data. Typically, this data gets transmitted to distant cloud data centers for analysis and storage, significantly impacting device performance and user experience. However, introducing the fog nodes at intermediate layers can alleviate the aforesaid issues by conducting most data processing operations locally and offloading only the resource-intensive tasks to the cloud. This paper proposes to showcase the efficacy of the hybrid fog-cloud-based framework for the smart refrigerator inventory management. Using a simulation tool, we evaluated the framework's efficiency against a conventional cloud-based system. The results demonstrate the clear advantages of our proposed framework with respect to network usage, power consumption and delay. Â© 2024 IEEE.
A Multi-Protocol Home Automation System Using Smart Gateway
(Springer, 2021) Chaudhary, S.K.; Yousuff, S.; Meghana, N.P.; Ashwin, T.S.; Guddeti, R.M.R.
Smart Home is one of the most established applications of the Internet of Things. Almost every equipment we use in our daily life—appliances, electric lights, electrical outlets, heating, and cooling systems-connected to a remotely controllable network, giving the user’s ability to remotely control and monitor the house, save energy without compromising on comfort and ultimately improve the quality of experience of staying in the house. We present a cost-effective system and address a major challenge that the industry faces today-Protocol Compatibility. To address the challenge, we make use of separate gateways/bridges for each network and an open-source home automation framework called OpenHAB, where each bridge links with a single master Wi-Fi gateway, providing a single window of control through an Application or a web interface for an integrated Smart Home. We integrate an elderly health monitoring device-Beehealth with OpenHAB; addressing the paramount need of a portable, accurate, and efficient health monitoring and fall detection device. We present two methods for fall detection, namely: threshold-based and Neural Network-based, with the latter resulting in 94% accuracy for fall detection. We evaluate the Smart Home devices on parameters like syncing time, battery life, recharge time, deployability, and cost. © 2020, Springer Science+Business Media, LLC, part of Springer Nature.
A Novel Fake Job Posting Detection: An Empirical Study andÂ Performance Evaluation Using ML andÂ Ensemble Techniques
(Springer Science and Business Media Deutschland GmbH, 2023) Srikanth, C.; Rashmi, M.; Ramu, S.; Guddeti, R.M.R.
Recently, everything can be accomplished online, including education, shopping, banking, etc. This technological advancement makes it easy for fraudsters to scam people online and acquire easy money. Numerous cyber crimes worldwide exist, including identity theft and fake job postings. Nowadays, many companies post job openings online, making recruitment simple. Consequently, fraudsters also post job openings online to obtain money and personal information from job seekers. In the proposed work, we aimed to decrease the frequency of such scams by using ensemble techniques such as AdaBoost, Gradient Boost, Stacking classifier, XgBoost, Bagging, and Random Forest to identify fake job postings from genuine ones. This paper proposes various featurization techniques such as Response coding with Laplace smoothing, Average Word2vec, and term frequency-inverse document frequency weighted Word2vec. We compared the performance of ensemble techniques with machine learning (ML) algorithms on publicly available EMSCAD dataset using accuracy and F1-score. Bagging classifier outperformed all the models with an accuracy of 98.85% and an F1-score of 0.88 on imbalanced dataset. On balanced dataset, XgBoost achieved 97.89% accuracy and 0.98 F1-score. From the experimental results, it is observed that a combination of ensemble and featurization techniques using Laplace smoothed Response coding and BoW stood superior to most of the state-of-the-art works on fake job posting detection. Â© 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
A novel hybrid algorithm for overlapping community detection in social network using community forest model and nash equilibrium
(Springer Verlag service@springer.de, 2019) Sarswat, A.; Guddeti, R.M.R.
Overlapping community detection in social networks is known to be a challenging and complex NP-hard problem. A large number of heuristic approaches based on optimization functions like modularity and modularity density are available for community detection. However, these approaches do not always give an optimum solution, and none of these approaches are able to clearly provide a stable overlapping community structure. Hence, in this paper, we propose a novel hybrid algorithm to detect the overlapping communities based on the community forest model and Nash equilibrium. In this work, overlapping community has been detected using backbone degree and expansion of the community forest model, and then a Nash equilibrium is found to get a stable state of overlapping community arrangement. We tested the proposed hybrid algorithm on standard datasets like Zachary’s karate club, football, etc. Our experimental results demonstrate that the proposed approach outperforms the current state-of-the-art methods in terms of quality, stability, and less computation time. © Springer Nature Singapore Pte Ltd. 2019
A novel real-time face detection system using modified affine transformation and Haar cascades
(Springer Verlag service@springer.de, 2019) Sharma, R.; Ashwin, T.S.; Guddeti, R.M.R.
Human Face Detection is an important problem in the area of Computer Vision. Several approaches are used to detect the face for a given frame of an image but most of them fail to detect the faces which are tilted, occluded, or with different illuminations. In this paper, we propose a novel real-time face detection system which detects the faces that are tilted, occluded, or with different illuminations, any difficult pose. The proposed system is a desktop application with a user interface that not only collects the images from web camera but also detects the faces in the image using a Haar-cascaded classifier consisting of Modified Census Transform features. The problem with cascaded classifier is that it does not detect the tilted or occluded faces with different illuminations. Hence to overcome this problem, we proposed a system using Modified Affine Transformation with Viola Jones. Experimental results demonstrate that proposed face detection system outperforms Viola–Jones method by 6% (99.7% accuracy for the proposed system when compare to 93.5% for Voila Jones) with respect to three different datasets namely FDDB, YALE and “Google top 25 ‘tilted face’” image datasets. © Springer Nature Singapore Pte Ltd. 2019
A novel sentiment analysis of social networks using supervised learning
(Springer-Verlag Wien michaela.bolli@springer.at, 2014) Anjaria, M.; Guddeti, R.M.R.
Online microblog-based social networks have been used for expressing public opinions through short messages. Among popular microblogs, Twitter has attracted the attention of several researchers in areas like predicting the consumer brands, democratic electoral events, movie box office, popularity of celebrities, the stock market, etc. Sentiment analysis over a Twitter-based social network offers a fast and efficient way of monitoring the public sentiment. This paper studies the sentiment prediction task over Twitter using machine-learning techniques, with the consideration of Twitter-specific social network structure such as retweet. We also concentrate on finding both direct and extended terms related to the event and thereby understanding its effect. We employed supervised machine-learning techniques such as support vector machines (SVM), Naive Bayes, maximum entropy and artificial neural networks to classify the Twitter data using unigram, bigram and unigram + bigram (hybrid) feature extraction model for the case study of US Presidential Elections 2012 and Karnataka State Assembly Elections (India) 2013. Further, we combined the results of sentiment analysis with the influence factor generated from the retweet count to improve the prediction accuracy of the task. Experimental results demonstrate that SVM outperforms all other classifiers with maximum accuracy of 88 % in predicting the outcome of US Elections 2012, and 68 % for Indian State Assembly Elections 2013. © 2014, Springer-Verlag Wien.
Adopting elitism-based Genetic Algorithm for minimizing multi-objective problems of IoT service placement in fog computing environment
(Academic Press, 2021) Natesha, B.V.; Guddeti, R.M.R.
Fog computing is an emerging computation technology for handling and processing the data from IoT devices. The devices such as the router, smart gateways, or micro-data centers are used as the fog nodes to host and service the IoT applications. However, the primary challenge in fog computing is to find the suitable nodes to deploy and run the IoT application services as these devices are geographically distributed and have limited computational resources. In this paper, we design the two-level resource provisioning fog framework using docker and containers and formulate the service placement problem in fog computing environment as a multi-objective optimization problem for minimizing the service time, cost, energy consumption and thus ensuring the QoS of IoT applications. We solved the said multi-objective problem using the Elitism-based Genetic Algorithm (EGA). The proposed approach is evaluated on fog computing testbed developed using docker and containers on 1.4 GHz 64-bit quad-core processor devices. The experimental results demonstrate that the proposed method outperforms other state-of-the-art service placement strategies considered for performance evaluation in terms of service cost, energy consumption, and service time. © 2021 Elsevier Ltd
Affective database for e-learning and classroom environments using Indian students’ faces, hand gestures and body postures
(Elsevier B.V., 2020) Ashwin, T.S.; Guddeti, R.M.R.
Automatic recognition of the students’ affective states is a challenging task. These affective states are recognized using their facial expressions, hand gestures, and body postures. An intelligent tutoring system and smart classroom environment can be made more personalized using students’ affective state analysis, and it is performed using machine or deep learning techniques. Effective recognition of affective states is mainly dependent on the quality of the database used. But, there exist very few standard databases for the students’ affective state recognition and its analysis that works for both e-learning and classroom environments. In this paper, we propose a new affective database for both the e-learning and classroom environments using the students’ facial expressions, hand gestures, and body postures. The database consists of both posed (acted) and spontaneous (natural) expressions with single and multi-person in a single image frame with more than 4000 manually annotated image frames with object localization. The classification was done manually using the gold standard study for both Ekman's basic emotions and learning-centered emotions, including neutral. The annotators reliably agree when discriminating against the recognized affective states with Cohen's ? = 0.48. The created database is more robust as it considers various image variants such as occlusion, background clutter, pose, illumination, cultural & regional background, intra-class variations, cropped images, multipoint view, and deformations. Further, we analyzed the classification accuracy of our database using a few state-of-the-art machine and deep learning techniques. Experimental results demonstrate that the convolutional neural network based architecture achieved an accuracy of 83% and 76% for detection and classification, respectively. © 2020 Elsevier B.V.
An Efficient AI and IoT Enabled System for Human Activity Monitoring and Fall Detection
(Institute of Electrical and Electronics Engineers Inc., 2024) Verma, N.; Mundody, S.; Guddeti, R.M.R.
Falls present a significant health risk, particularly among the elderly, necessitating reliable wearable fall detection systems. This paper introduces an advanced AI-powered system that integrates Generative Adversarial Networks (GANs) for synthetic data augmentation and Convolutional Neural Networks (CNNs) for robust fall detection and daily activity recognition. The primary challenge in developing effective fall detection systems lies in the scarcity and diversity of real-world fall data. This paper addresses this challenge innovatively by employing a GAN trained on datasets of authentic fall events to generate synthetic data. This augmentation strategy significantly expands the training dataset, enhancing the model's capacity to generalize across various fall scenarios and daily activities. The system leverages a specialized 1D CNN architecture designed for processing accelerometer and gyroscope readings obtained from wearable devices, enabling precise feature extraction to distinguish subtle differences between falls and routine movements. The evaluation results demonstrate a notable advancement by achieving a superior accuracy of 99 % for fall detection while minimizing false positives. The developed CNN model can also classify 15 kinds of falls and 19 types of daily life activities. Â© 2024 IEEE.
An Ensemble Deep Learning Approach for Emotion Monitoring System in Online Examinations
(Institute of Electrical and Electronics Engineers Inc., 2024) Bhardwaj, S.; Ramu, S.; Guddeti, R.M.R.
Around the world, a large number of students experience difficult life situations that have an effect on their emotional and mental health and, ultimately, their academic performance in examinations (exams in short). Emotions have an effect on a student's motivation, focus, and memory, and finally how they perform in exams. An Emotion Monitoring System could be really helpful for understanding how students are feeling during exams and how it affects their overall performance. The contribution of this paper involves in designing a novel facial emotion tracking system which can be used for analyzing facial expressions in real-time thus providing timely emotional support during exams. In this work, we utilized five pretrained deep learning models, namely: DenseNet-121, MobileNetV2, EfficientNet-B0, Inception-V3 and Xception - to classify emotions on processed Emoset dataset. Further, we developed an ensemble model by fusing aforementioned two top-performing deep learning models, thus harnessing the strengths of both models. From the results it can be inferred that ensemble model outperforms the individual pretrained models giving an accuracy of 98.67%. The superior performance of the ensemble models makes it an ideal choice for implementing emotion recognition in real-time applications like Emotion Monitoring in exams. Â© 2024 IEEE.
An Optimized Question Classification Framework Using Dual-Channel Capsule Generative Adversarial Network and Atomic Orbital Search Algorithm
(Institute of Electrical and Electronics Engineers Inc., 2023) Revanesh, M.; Rudra, B.; Guddeti, R.M.R.
The advancement in education has emphasized the need to evaluate the quality of the examination questions and the cognitive levels of students. Many educational institutions now acknowledge Bloom's taxonomy-based students' cognitive levels evaluating subject-related learning. Therefore, in this paper, a novel optimized Examination Question Classification framework, referred to as QC-DcCapsGAN-AOSA, is proposed by combining the Dual-channel Capsule generative Adversarial Network (DcCapsGAN) with Atomic Orbital Search Algorithm (AOSA) for preprocessing a real-time online dataset of university examination questions, thus identify the key features from the raw data using Term Frequency Inverse Document Frequency (TF-IDF) and finally classifying the examination questions. Atomic Orbital Search Algorithm is used to fine-tune the parameters' weights of the DcCapsGAN, and then uses these weights to categorize questions as Knowledge Level, Comprehension Level, Application Level, Analysis Level, Synthesis Level, and Evaluation Level. Experimental results demonstrate the superiority of the proposed method (QC-DcCapsGAN-AOSA) when compared to the state-of-the-art methods such as QC-LSTM-CNN and QC-BiGRU-CNN with an accuracy improvement of 23.65% and 29.04%, respectively. © 2013 IEEE.
Automated Parking System in Smart Campus Using Computer Vision Technique
(Institute of Electrical and Electronics Engineers Inc., 2019) Banerjee, S.; Ashwin, T.S.; Guddeti, R.M.R.
In today's world we need to maintain safety and security of the people around us. So we need to have a well connected surveillance system for keeping active information of various locations according to our needs. A real-time object detection is very important for many applications such as traffic monitoring, classroom monitoring, security rescue, and parking system. From past decade, Convolutional Neural Networks is evolved as a powerful models for recognizing images and videos and it is widely used in the computer vision related work for the best and most used approach for different problem scenario related to object detection and localization. In this work, we have proposed a deep convolutional network architecture to automate the parking system in smart campus with modified Single-shot Multibox Detector (SSD) approach. Further, we created our dataset to train and test the proposed computer vision technique. The experimental results demonstrated an accuracy of 71.2% for the created dataset. Â© 2019 IEEE.
Automatic detection of students’ affective states in classroom environment using hybrid convolutional neural networks
(Springer, 2020) Ashwin, A.; Guddeti, R.M.R.
Predicting the students’ emotional and behavioral engagements using computer vision techniques is a challenging task. Though there are several state-of-the-art techniques for analyzing a student’s affective states in an e-learning environment (single person’s engagement detection in a single image frame), a very few works are available for analyzing the students’ affective states in a classroom environment (multiple people in a single image frame). Hence, in this paper, we propose a novel hybrid convolutional neural network (CNN) architecture for analyzing the students’ affective states in a classroom environment. This proposed architecture consists of two models, the first model (CNN-1) is designed to analyze the affective states of a single student in a single image frame and the second model (CNN-2) uses multiple students in a single image frame. Thus, our proposed hybrid architecture predicts the overall affective state of the entire class. The proposed architecture uses the students’ facial expressions, hand gestures and body postures for analyzing their affective states. Further, due to unavailability of standard datasets for the students’ affective state analysis, we created, annotated and tested on our dataset of over 8000 single face in a single image frame and 12000 multiple faces in a single image frame with three different affective states, namely: engaged, boredom and neutral. The experimental results demonstrate an accuracy of 86% and 70% for posed and spontaneous affective states of classroom data, respectively. © 2019, Springer Science+Business Media, LLC, part of Springer Nature.
Bat-termite: A novel hybrid bio inspired routing protocol for mobile ad hoc networks
(Inderscience Publishers, 2014) Manjappa, M.; Guddeti, R.M.R.
In this paper, the authors have proposed a novel hybrid bio-inspired routing protocol for Mobile Ad Hoc Networks (MANETs), referred to as bat-termite, by combining the unique features of both social insect termites and mammals bats. The primary objective of the proposed work is to design an adaptive routing protocol for MANETs based on the hill building nature of the termites. The secondary objective of the proposed work is to improve the backup route maintenance of the proposed algorithm using the echo-location feature of the bats. The proposed bat-termite algorithm exhibits superior routing features such as quick route discovery, high robustness with efficient management of multiple routes and rapid route repair. The bat-termite algorithm is simulated in NS-2 and the simulation results are compared with the bio-inspired (termite and D-Termite) and non bio-inspired (AODV and AOMDV) routing protocols from the performance evaluation point of view. Copyright © 2014 Inderscience Enterprises Ltd.
Bio-Inspired Hyperparameter Tuning of Federated Learning for Student Activity Recognition in Online Exam Environment
(Multidisciplinary Digital Publishing Institute (MDPI), 2024) Ramu, R.; Prasad, N.; Guddeti, R.M.R.; Mohan, B.R.
Nowadays, online examination (exam in short) platforms are becoming more popular, demanding strong security measures for digital learning environments. This includes addressing key challenges such as head pose detection and estimation, which are integral for applications like automatic face recognition, advanced surveillance systems, intuitive human–computer interfaces, and enhancing driving safety measures. The proposed work holds significant potential in enhancing the security and reliability of online exam platforms. It achieves this by accurately classifying students’ attentiveness based on distinct head poses, a novel approach that leverages advanced techniques like federated learning and deep learning models. The proposed work aims to classify students’ attentiveness with the help of different head poses. In this work, we considered five head poses: front face, down face, right face, up face, and left face. A federated learning (FL) framework with a pre-trained deep learning model (ResNet50) was used to accomplish the classification task. To classify students’ activity (behavior) in an online exam environment using the FL framework’s local client device, we considered the ResNet50 model. However, identifying the best hyperparameters in the local client ResNet50 model is challenging. Hence, in this study, we proposed two hybrid bio-inspired optimized methods, namely, Particle Swarm Optimization with Genetic Algorithm (PSOGA) and Particle Swarm Optimization with Elitist Genetic Algorithm (PSOEGA), to fine-tune the hyperparameters of the ResNet50 model. The bio-inspired optimized methods employed in the ResNet50 model will train and classify the students’ behavior in an online exam environment. The FL framework trains the client model locally and sends the updated weights to the server model. The proposed hybrid bio-inspired algorithms outperform the GA and PSO when independently used. The proposed PSOGA not only outperforms the proposed PSOEGA but also outperforms the benchmark algorithms considered for performance evaluation by giving an accuracy of 95.97%. © 2024 by the authors.
Communication and computation optimization of concurrent kernels using kernel coalesce on a GPU
(John Wiley and Sons Ltd, 2015) Bayyapu, B.; Guddeti, R.M.R.; Raghavendra, P.S.
General purpose computation on graphics processing unit (GPU) is rapidly entering into various scientific and engineering fields. Many applications are being ported onto GPUs for better performance. Various optimizations, frameworks, and tools are being developed for effective programming of GPU. As part of communication and computation optimizations for GPUs, this paper proposes and implements an optimization method called as kernel coalesce that further enhances GPU performance and also optimizes CPU to GPU communication time. With kernel coalesce methods, proposed in this paper, the kernel launch overheads are reduced by coalescing the concurrent kernels and data transfers are reduced incase of intermediate data generated and used among kernels. Computation optimization on a device (GPU) is performed by optimizing the number of blocks and threads launched by tuning it to the architecture. Block level kernel coalesce method resulted in prominent performance improvement on a device without the support for concurrent kernels. Thread level kernel coalesce method is better than block level kernel coalesce method when the design of a grid structure (i.e., number of blocks and threads) is not optimal to the device architecture that leads to underutilization of the device resources. Both the methods perform similar when the number of threads per block is approximately the same in different kernels, and the total number of threads across blocks fills the streaming multiprocessor (SM) capacity of the device. Thread multi-clock cycle coalesce method can be chosen if the programmer wants to coalesce more than two concurrent kernels that together or individually exceed the thread capacity of the device. If the kernels have light weight thread computations, multi clock cycle kernel coalesce method gives better performance than thread and block level kernel coalesce methods. If the kernels to be coalesced are a combination of compute intensive and memory intensive kernels, warp interleaving gives higher device occupancy and improves the performance. Multi clock cycle kernel coalesce method for micro-benchmark1 considered in this paper resulted in 10-40% and 80-92% improvement compared with separate kernel launch, without and with shared input and intermediate data among the kernels, respectively, on a Fermi architecture device, that is, GTX 470. A nearest neighbor (NN) kernel from Rodinia benchmark is coalesced to itself using thread level kernel coalesce method and warp interleaving giving 131.9% and 152.3% improvement compared with separate kernel launch and 39.5% and 36.8% improvement compared with block level kernel coalesce method, respectively. © 2014 John Wiley & Sons, Ltd.
Deep learning-based multi-view 3D-human action recognition using skeleton and depth data
(Springer, 2023) Ghosh, S.K.; Rashmi, M.; Mohan, B.R.; Guddeti, R.M.R.
Human Action Recognition (HAR) is a fundamental challenge that smart surveillance systems must overcome. With the rising affordability of capturing human actions with more advanced depth cameras, HAR has garnered increased interest over the years, however the majority of these efforts have been on single-view HAR. Recognizing human actions from arbitrary viewpoints is more challenging, as the same action is observed differently from different angles. This paper proposes a multi-stream Convolutional Neural Network (CNN) model for multi-view HAR using depth and skeleton data. We also propose a novel and efficient depth descriptor, Edge Detected-Motion History Image (ED-MHI), based on Canny Edge Detection and Motion History Image. Also, the proposed skeleton descriptor, Motion and Orientation of Joints (MOJ), represent the appropriate action by using joint motion and orientation. Experimental results on two datasets of human actions: NUCLA Multiview Action3D and NTU RGB-D using a Cross-subject evaluation protocol demonstrated that the proposed system exhibits the superior performance as compared to the state-of-the-art works with 93.87% and 85.61% accuracy, respectively. © 2022, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.
Depth-Based Selective Blurring in Stereo Images Using Accelerated Framework
(3D Research Center 3drc@kw.ac.kr, 2014) Mukherjee, S.; Guddeti, R.M.R.
Abstract: We propose a hybrid method for stereo disparity estimation by combining block and region-based stereo matching approaches. It generates dense depth maps from disparity measurements of only 18 % image pixels (left or right). The methodology involves segmenting pixel lightness values using fast K-Means implementation, refining segment boundaries using morphological filtering and connected components analysis; then determining boundaries’ disparities using sum of absolute differences (SAD) cost function. Complete disparity maps are reconstructed from boundaries’ disparities. We consider an application of our method for depth-based selective blurring of non-interest regions of stereo images, using Gaussian blur to de-focus users’ non-interest regions. Experiments on Middlebury dataset demonstrate that our method outperforms traditional disparity estimation approaches using SAD and normalized cross correlation by up to 33.6 % and some recent methods by up to 6.1 %. Further, our method is highly parallelizable using CPU–GPU framework based on Java Thread Pool and APARAPI with speed-up of 5.8 for 250 stereo video frames (4,096 × 2,304). © 2014, 3D Research Center, Kwangwoon University and Springer-Verlag Berlin Heidelberg.